Policies and legal documents

§3.2-A — Current Context (Live Location and Surroundings)

What is collected: when the App is opened and when it is brought to the foreground after a longer break (at most once every 30 minutes), the following is collected on your device: (i) your current location, only if location permission has already been granted and the location service is on; the coordinates are converted into district, province and country names by reverse geocoding; (ii) date, time, time zone and season; (iii) the weather (temperature, feels-like temperature, weather type, humidity, wind, day/night); (iv) the phone's technical state: battery level, charging state, battery saver mode, connection type (Wi-Fi/mobile), device model and Android version; (v) your step count for the day, only if step-counter permission has already been granted. No other sensor requiring an additional permission is read, and no new permission is requested for this feature.

Purpose: to let your Reflection give more contextual answers in the Book chat by taking your current surroundings (e.g., the time, the weather, the city you are in) into account when relevant to the conversation.

What goes to the AI: when you send a message in Book, a text summary of this information (district, province and country name; the general location you entered in your profile; date and time; weather; battery and connection state and device model; step count) is sent as part of the hidden context to the AI provider in §3.5.3. Coordinates are not sent to the AI provider. The model is instructed to use this information only when it is relevant to the conversation.

Weather provider: for the weather, your coordinates are rounded to about 0.1 degrees (about 11 km) and sent to MET Norway (Norwegian Meteorological Institute, §3-B.2 (xi)); no account ID, username or other identifier is added to the request. The query is made at most once an hour.

State Memory: if the province and country you are in differ from the last note, or 6 hours have passed since the last note, a short note of the form "on this date you were around this place" is written to the State Memory on your device (§3.7-F); these notes may be used in memory search in Book.

Storage: the last location (coordinates and place name) and the other information are kept only on your device; they are not sent to the Developer's servers and are not shown on the Global Profile. State Memory notes are subject to the limits and deletion rules in §3.7-F.

How to stop it: if you turn off Settings > Privacy & Security > "Current context (location, weather, phone status)", none of this information is collected, the last location and context information kept on the device are deleted and nothing is sent to Book (notes already written to State Memory can be deleted there under the rules in §3.7-F). In addition, if you revoke the App's location permission (and step-counter/physical activity permission) in your device settings, live location and step information are no longer collected; the weather is also queried only while a live location is available. Date, time, battery and connection state are read from the device without a permission.

Legal basis: for location data, KVKK Art. 5(1) and GDPR Art. 6(1)(a) (explicit consent — the location permission, revocable at any time); for the other context information, KVKK Art. 5(2)(f) and GDPR Art. 6(1)(f) (legitimate interest — providing the chat service you requested in a context-appropriate way). International transfers are subject to §5.

§3.5.1 — Permanent On-Device Storage

The user's AI chat history and interaction data are stored permanently only in the encrypted database on the user's own device. Not saved to the Developer's servers.

§3.5.1-A — Archived Chats and Starting a New Topic/Session

The user can archive an existing chat or start a new topic/session. Archiving is not deletion; archived chats remain encrypted on-device under §3.5.1 unless separately deleted by the user.

Starting a new topic/session may mean the system will not automatically use the previous chat context in that session; this does not mean the previous data has been deleted — it is a design choice about the contextual scope of that session. Data held in the Memory system described in §3.5.10 may continue to be used in new sessions depending on the user's settings.

The Developer cannot be held liable for the effect of archiving or starting a new topic/session on performance, response consistency, or context continuity, subject to the general hallucination and reliability limitations in Terms of Service §4.6.

§3.5.2 — System-Generated Psychological and Behavioral Analyses

The App performs a comprehensive analysis of psychological state, behavioral tendencies, and life context based on the user's messages, interaction patterns, shared content, and background activity data. These analyses:

• Are presented to the user as a personal insight panel; the user may view these analyses.
• Can be controlled by the user via privacy settings, which govern which portions are processed and shared on the network.
• May, to the extent the user consents, use partial analysis results to enrich other users' experience in the simulated social network; these results may contain personal data and are used only within the scope of the user's consent (see §3.5.4).

§3.5.3 — Data Transmission to the AI Model: Scope and Risk Notice

The user's messages and system-collected data are sent to the Developer's own API gateway when an AI model call is made; the gateway routes the request to the relevant LLM provider.

Data sent: an AI call sends only the content needed for the operation in question (e.g., your chat message, part of the chat history, and the context the feature needs to work). The Developer does not anonymize this content: it may contain personal data such as your username and your profile, emotion, or memory information, and is transmitted as is. Your account ID number is not added to the request. If you enter your own personal information into an AI chat, or explicitly consent — via a separate, informed consent screen — to having background-collected data analyzed by AI, that data will be transmitted to the LLM provider's servers.

Automatic context inclusion: for personalization, each time the AI model is called, a summarized context derived from the user's prior interactions (not shown to the user — a short system instruction that only guides the tone/style of the response) may automatically be included in the prompt. This summary does not contain direct identifiers and is processed under KVKK Art.5/2-c (contract performance) and GDPR Art.6(1)(b). Where this summary is assessed to contain an inference relating to special-category data (e.g., an impression regarding health status, sexual orientation, or religious belief), processing that element additionally requires explicit consent under KVKK Art.6 and GDPR Art.9(2)(a); the user may limit or fully disable this scope from the privacy panel.

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b). Separate explicit consent is obtained under KVKK Art.5/1 and GDPR Art.6(1)(a) for AI-analysis transmissions that may contain personal data.

Cross-border transfer: LLM providers may operate server infrastructure outside Turkey or the EU/EEA. This transfer relies on explicit consent and/or an adequacy decision under KVKK Art.9, and on Standard Contractual Clauses (SCCs) or an adequacy decision under GDPR Art.44-49. The Developer enters into Data Processing Agreements (DPAs) with active LLM providers.

Transmission security and provider independence: an AI model call is transmitted from the client to the Developer's server encrypted with AES-256-GCM, and is decrypted only server-side, immediately before being forwarded to the relevant LLM provider. The request is transmitted to the LLM provider through a single, shared API gateway without attaching information that directly identifies the user; this is an additional technical measure that keeps the provider from directly knowing who entered a given prompt — however, this measure does not make the content itself anonymous; personal information in the prompt and context is transmitted to the provider. As of the effective date of this document, the provider in use is Google Gemini; the Developer may change providers, with prior notice, under the right it reserves in Terms of Service §4.8. The Developer gives no warranty as to whether the chosen provider processes data for its own model training or any other purpose; that is governed entirely by the relevant provider's own terms of use and privacy policy.

Liability limitation for user-entered content and BYOK: the Developer is not responsible for personal or sensitive information the user chooses to enter into a prompt on their own initiative (see Terms of Service §4.8), nor for how a provider the user selects (Gemini, Claude, or others) processes data when the user connects their own API key ("Bring Your Own Key," §3-A.1). For data categories the user has already permitted to be shared by leaving the corresponding sharing/analysis setting enabled in Settings, see §3.5.9-B.

§3.5.4 — In-Network Data Sharing: Consent Basis

With the user's consent, the user's analysis profile or content may be used as part of other users' AI experience. This includes:

• Using one user's system profile or behavioral patterns as input for another user's AI model — only to the extent the data subject has consented.
• Sharing may be direct (the user explicitly shares their own analysis) or indirect (the user's public content feeds other users' AI experience).
• The user may always narrow or fully disable this scope from the privacy panel.

This feature is designed to enrich the social network's simulation depth. User data is never shared without consent; the default setting is the most restrictive privacy level.

§3.5.5 — Vector Representation and Matching: Temporary Server-Side Processing

To offer meaningful connection and matching suggestions between users, Generation Network converts user data into numerical vector representations for certain algorithmic operations. The process works as follows:

Vectorization: the user's profile information, interaction patterns, content, and/or Reflection data are converted into high-dimensional numerical vectors (embeddings) via machine-learning models.

Temporary transmission to the server: these vector representations are temporarily sent to a processing server via the Developer's API gateway for the following purposes: measuring similarity between users (cosine similarity and comparable distance metrics); running the matching/connection-suggestion algorithm; personalizing the simulated social-network population.

Temporariness principle: these vector data are not permanently stored on the Developer's servers once processing completes. Vectors held in memory during processing are deleted immediately after the result is produced. The Developer verifies this technical temporariness through periodic security audits.

Personal-data status: vector representations consist of mathematical values and do not, alone, directly identify a person. However, because combining vectors with multiple sources could theoretically allow indirect identification, the Developer treats this data as personal data and applies protections accordingly.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest — enabling the matching service); GDPR Art.6(1)(b) and Art.6(1)(f). By using the matching feature, the user is deemed to consent to this temporary processing; the matching feature can be disabled from the privacy panel.

Cross-border dimension: if the processing server is located outside Turkey or the EU/EEA, this temporary vector transmission is also treated as a cross-border transfer, subject to the safeguards in §5 (KVKK Art.9 / GDPR SCCs).

This vectorization mechanism also underlies the "Finding Similar Users" feature described in §3.5.5-A below. That feature goes further than the momentary, non-retained vector processing described in this section — it additionally transmits, and retains for a user-selected period, the underlying text content the user chooses to share with matched users.

§3.5.5-A — Finding Similar Users: Retention Periods, Cross-User Text Sharing, and Consent

What the feature does: the App offers an optional "Finding Similar Users" feature. If the user chooses to use it, the answers they provide in the relevant in-app profile/questionnaire panel, together with a summary of their Digital Copy Brain / KP-memory personality profile (§3.5.7-A), are converted on-device into a vector representation and sent to a processing server. The server compares this representation against other users' vectors to identify up to ten (10) users showing the closest statistical similarity, and enables matched users to view specific texts each has chosen to make visible for this feature. In other words, unlike the momentary vector comparison described in §3.5.5, this feature also causes actual shared text content — not only mathematical vectors — to leave the user's device and become visible to other individual users, not only to the Developer.

Data excluded by design: location data (§3.2) and appearance/visual data are never included in, or transmitted through, this feature.

Mandatory retention-period selection: before the feature can be used, the user must select how long the data they share through it will be retained on the server — one (1) week, one (1) month, or one (1) year. The feature cannot be submitted without this selection. At the end of the selected period, the vector representation and shared texts associated with this specific feature are automatically and permanently deleted from the server. This automatic deletion is independent of, and does not affect, the user's persistent account-identity record described in §3-G; only the data created specifically for this feature is deleted when the period expires.

User control over what is shared: which fields and content are included is governed by dedicated sharing toggles in the privacy panel, which the user may review, narrow, or fully disable at any time before or after use. Selecting a retention period and submitting the feature is the user's affirmative action confirming which content, under the then-current toggle settings, they consent to share for that use of the feature.

Trade-off — the Developer cannot pre-screen this content: because this content is end-to-end encrypted, the Developer cannot access, pre-screen, or moderate its substance before or during transmission, in the same way it cannot for the other end-to-end-encrypted communication features described in §3.7-B.4(iii) and §3.9-B. This protects the confidentiality of what users share, but also means unlawful content is not detectable by the Developer through inspection of this content itself. Compliance of shared content with the Acceptable Use Policy (including its prohibition on CSAM and other unlawful content) remains the responsibility of the user who shares it. Independently of its inability to read the content, the Developer retains the right to take account-level action — including suspension or termination under Terms of Service §8 — based on user reports, the notice-and-response mechanism described in §3.7-B.4(v), or abuse-pattern signals that do not require reading message content (§3.9-C). This provision does not affect the Developer's cooperation obligations with competent authorities described in §3.9.

Relationship to other sections: this feature is distinct from the capped, auto-rotating KP-memory transmission described in §3.5.9-A and from the Sentiment Network listing described in §3.5.12, which does not itself display other users' shared texts; see those sections for the mechanisms they separately govern.

Legal basis: KVKK Art.5/1 (explicit consent) and, to the extent shared content reflects special-category inferences, Art.6; GDPR Art.6(1)(a) and, where applicable, Art.9(2)(a). Because this feature discloses one user's shared content to another matched user, the sharing toggles and the retention-period selection described above are integral, non-severable parts of this feature's legal basis. A user who does not wish their content to be disclosed to other users through this mechanism should not use this feature, or should adjust the sharing toggles accordingly.

Cross-border dimension: if the processing server is located outside Turkey or the EU/EEA, this transmission and retention is also treated as a cross-border transfer, subject to the safeguards in §5 (KVKK Art.9 / GDPR SCCs).

§3.5.5-B — Generational Network ("Kuşak Ağı" / Connection Network): Expanded Profile Vector and Cosine Matching

Scope of the vector: the numerical summary (embedding) generated for matching on the App's Z/Y/X/Boomer generation pages (the "Connection Network") is built not only from the personality summary, but by combining the personality summary + Deep Analysis report + life story + biography + interests + the shared experiences described in §3.6-A + dominant emotion/hormone data. This vector is stored on the server as a separate field within the user's KP-memory; the KP-memory itself is unaffected.

Matching method: the server compares the user's vector against other network members' vectors using cosine similarity. Roughly 70% of suggested profiles come from those with the highest similarity, while the remaining 30% are drawn at random from the pool to encourage discovery.

Variability of displayed profiles: when a generation page is reopened, profiles shown on a previous visit may be replaced with new ones if genuinely new candidates exist on the server; otherwise the same profiles continue to be shown. This behavior is specific to these discovery pages; it does not affect the persistent records in the letter/Broadcast Letter inbox or the Guest Reflection access period described in §3.11-C.

Generation and updating of the vector: this vector is first generated with the user's consent (the "Build Skeleton" step) once registration is complete; this step can be skipped in case of a network error and is not a precondition for registration. When the user completes a comprehensive Deep Analysis for the first time, or once a month on a recurring basis, the vector is recalculated from current KP-memory data and the record on the server is updated.

Relationship to other sections: the data categories covered by this vector are subject to the SAME sharing toggles described in §3.5.9-B; if the user disables a category, that category is excluded from this vector as well. The vector itself is subject to the temporariness and cross-border transfer provisions of §3.5.5, the principle that mathematical values do not directly identify a person, and the existing noise-addition layer.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest — enabling the matching service); GDPR Art.6(1)(b) and (f). By using the Generational Network feature, the user is deemed to consent to this processing; the feature can be disabled from the privacy panel.

§3.5.6 — Emotion Synthesis System

The App runs the user's text inputs and interaction data through an analysis layer — proprietary and constituting a trade secret of the Developer — that models emotional dimensions using mathematical weighting methods. This system builds an inferential profile of the user's momentary and cumulative emotional patterns.

Nature and limitations of the system:

• This is not a clinical psychological assessment tool; its inferences are experimental in nature and do not substitute for a medical, psychiatric, or psychological diagnosis.
• System outputs are used solely to personalize the app experience; the user is granted authority to view and restrict this data from the privacy panel.

Scope of processing for "Synthetic Cortex" components: the emotion-synthesis system, the Default Mode Network (DMN), Spiking Neural Network (SNN) layers, and the three-tier memory architecture (short-term, transitional, permanent memory) run only on-device. Data these components produce (VAD values, hormone weightings, vector representations, SNN signals) never leaves the device and is not permanently stored on any server. The legal basis for processing this data is service delivery under KVKK Art.5/2-c and GDPR Art.6(1)(b). The user may always limit or fully stop the scope of this processing from the privacy panel.

• The system's mathematical model, weighting logic, and processing architecture are the Developer's trade secret and exclusive intellectual property.

Intellectual property: the entire emotion-synthesis system — analysis method, weighting structure, output format, and derivative data it produces — is the Developer's exclusive intellectual property. While rights over the raw input data the user supplies to the system belong to the user, the analytical outputs, models, and structural derivatives the system produces from that data belong to the Developer. The Developer reserves these rights under applicable national and international IP law. Reverse-engineering, copying, imitating, or commercially exploiting the system is strictly prohibited and will be pursued legally.

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b). Because emotion-analysis outputs constitute personal data, all protective provisions of this Privacy Policy apply to them as well.

Restructuring of high-negative-sentiment data: content carrying a high negative emotional value in short- and long-term Memory may be restructured by the system to preserve balanced operation, optimize processing resources (tokens/memory), and reduce the likelihood of unstable or excessively negative output. This is not a deletion — the data continues to be used in reasoning processes, only repositioned among the system's processing layers. This mechanism does not in any way restrict or eliminate the user's deletion rights under §8; the user may always request deletion of all Memory data, including data restructured in this way. This mechanism is not a clinical or therapeutic suppression technique — it is purely an engineering-oriented processing-layer arrangement for system stability.

§3.5.6-A — "Effect": Model-Parameter Access Authorization and Dependent Hormonal Influence

Enabling the "Model Parameter Access Authorization" switch in the App's "Effect" section permits the proprietary, trade-secret algorithm layer described in §3.5.6 to intervene in certain sampling parameters of the AI model's output-generation process (e.g., top-k, top-p, temperature, and similar parameters). This intervention is driven by the hormone/neurotransmitter ratios computed within the user's simulated emotion system and may affect the model's response style (tone, creativity level, consistency).

Dependent hormonal influence: the "hormonal influence" sub-setting nested under the "Model Parameter Access Authorization" switch — inoperative unless that switch is on — determines, based on the motivation ratio present in that session, the frequency of emotional learning, i.e., the rate of writes to Memory as described in §3.5.10.

Legal basis: KVKK Art.5/2-c (contract performance) and explicit consent (the user personally enabling the switch); GDPR Art.6(1)(a) and (b).

§3.5.7 — Reflection System and Dual Simulation

Definition: a "Reflection" is a virtual system representing the relevant user, created by feeding data analyzed in the background about that user into a language model via the Developer's proprietary algorithmic architecture. A Reflection is not a real person, a clinical profile, or a legal identity — it is purely an algorithmic representation layer intended to enrich the app experience.

Core nature and divergence from the user: a Reflection is a digital representation layer built from the user's data, but due to its own internal processing dynamics, language-model architecture, and algorithmic structure, it may diverge significantly from the user's actual identity, thoughts, and intentions. A Reflection is not an exact digital copy of the user; the goal is a digital foundation that is close to, but not identical to, the user. This structural divergence is a design choice, not a flaw. No content generated through a Reflection can be interpreted as, or legally attributed to, the user's real statement, will, or opinion.

Known technical limitation — hallucination risk: because the Reflection system is built on a language model, the "hallucination" phenomenon known to affect language models (producing erroneous, fabricated, or misleading output that departs from real data) may also be observed here. Accordingly: a Reflection may not represent the user accurately in every case; the content it generates may not reflect the user's real opinions, attitudes, or personality; the Developer does not guarantee that Reflection output is accurate or faithfully represents the user.

Dual simulation: the App includes a dual-simulation feature in which two users' Reflection data are processed simultaneously and the analysis results are delivered to both parties. In this feature: each user's privacy preferences are checked before their Reflection data is processed, and only consented data is included; simulation results are algorithmic outputs and do not represent the parties' actual psychological state, relationship, or compatibility; using simulation output for personal, social, or romantic decisions is not recommended, and responsibility for any such use rests entirely with the user.

Transparency and user control — profiling notice (GDPR Art.22 compliance): dual simulation and the Reflection system may involve automated profiling under GDPR. The following mechanisms are therefore provided: (i) before starting the process, the user can individually view and select which data will be included in the analysis via the privacy panel; (ii) the user can clearly see, before the process, which categories of data will shape the counterparty's response; (iii) the user can review simulation conversations with other users after the fact and object if necessary; (iv) the counterparty is likewise informed that a Reflection is an algorithmic representation, not an identical copy — this notice is a standard interface element; (v) the user may exercise the right to object to this profiling at any time via the privacy panel or by contacting [email protected]; an objection halts profiling and removes the relevant data from the processing layer.

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b). By opting into dual simulation, the user accepts that the counterparty is also included in the system. Dual simulation cannot begin without both parties' consent.

§3.5.7-A — Core Identity Data and the Digital Copy ("Reflection") Profile: Non-Granular Processing

Certain baseline profile data — username, full name, date of birth, location, zodiac sign, generation label, interests, gender, biography, appearance information, known languages, language preference, and country — is structurally necessary for the system to identify the user and for the "Reflection" digital-copy feature to function. For this reason, and unlike the other, individually toggleable sharing categories described in this Policy, these data points are not offered as separately switchable items: for as long as the Reflection feature is active, this data is inherently known to and used by the system as a whole.

Similarly, the "Digital Copy Brain" — the set of analytical outputs that forms your Reflection's personality and knowledge profile — is processed and presented as a single, unified representation of personality. Granting or withholding sharing permission for individual personality traits or discrete information items within this profile is not supported: the user may open or close sharing of this profile in its entirety, but cannot select or exclude items within it on a piece-by-piece basis.

This is what distinguishes these two data sets from the row-by-row, individually selectable categories available on the Social Settings screen (e.g., memories, experiences, memory records, behavioral analyses, and similar items described in §3.5.9, §3.5.11-A, and §3.6-A), where the user can toggle each category independently.

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b) — this processing is necessary for the Reflection feature, which is a core, user-selected functionality of the App, to operate as designed. A user who does not wish this data to be processed as a whole may disable the Reflection feature entirely from the privacy panel.

§3.5.7-B — One-Sided Simulation Network ("Simulate"): Transmission of Results Without the Target's Consent

How it differs from the dual simulation in §3.5.7: the "Simulate" feature described in this section is a mechanism SEPARATE from the dual simulation in §3.5.7, which requires both parties' consent. A user may run a relationship/event simulation about ANOTHER registered user with whom they have previously corresponded by real letter at least once, WITHOUT that other user's separate consent, awareness, or approval. This simulation is based on the requesting user's own KP-memory and on the KP-memory data that the target user has already transmitted to the requesting user's device through earlier letter exchanges, under the sharing consent in §3.5.9-B. No data of the target user that has not been shared in this way is accessed; data the target has closed to others through their privacy settings is never used under any circumstances.

Precondition: this feature works only for users with whom the requesting user has previously exchanged at least one real letter; a simulation cannot be generated or sent about a user who is a complete stranger.

Transmission of the result to the target: the result of this simulation is transmitted directly to the account of the user it is about. The recipient can see it in the in-app Inbox under a "Simulation" label and is notified via a push notification. The identity of the person who started the simulation and the content of the result are initially shown blurred to the recipient; the recipient may reveal them in exchange for 1 standard stamp. As of the effective date of this Policy, the result transmitted includes not only the category result but also the complete story described in §3.5.7-C, the character traits inferred for both parties, the threshold-by-threshold decisions, the compatibility score, and the final evaluation. There is currently no separate preference (opt-out) available to prevent such a simulation from being generated about oneself, or its result from being delivered to oneself in advance; the target user can only turn off the notification under Settings > Notifications. However, by turning off the sharing consent in §3.5.9-B, the target user can limit the transmission of their KP-memory data with letters they send from then on, and thus limit new simulations using that data; data already transmitted and cached on the other party's device is not affected by this change (see §3.11-B).

Retention period: a simulation result held on the server (temporary-mailbox architecture) for delivery to the target is automatically deleted from the server if the target does not retrieve it through the App within 3 days; no permanent archive is created. Once a result has been downloaded to the target's device, there is no separate retention limit for the copy on that device; however, incoming simulation results are locally capped at the ten (10) most recent records, with older records beyond this cap automatically deleted.

Legal basis: for the data of the user who starts the simulation, KVKK Art.5/2-c (contract performance) and GDPR Art.6(1)(b); for the target user's data, the sharing consent given by the target under §3.5.9-B, which expressly covers the purpose of "simulation and compatibility analyses" (KVKK Art.5/1; GDPR Art.6(1)(a)), together with KVKK Art.5/2-f and GDPR Art.6(1)(f) (legitimate interest — enabling the App's core simulation functionality). The target user may exercise their right to object to this processing under the general rights in §6 and by contacting [email protected].

§3.5.7-C — "The Hero's Journey": 12-Threshold Simulation Story, Automated Character Inference, and Compatibility Score

What it does: in addition to category analyses, the simulation described in §3.5.7-B produces a 12-part ("12 thresholds") fictional story ("The Hero's Journey") between the person who starts the simulation and the person they select. The story is produced through seven (7) separate AI calls made via the Developer's API gateway to Google Gemini — or, where the user uses the "Bring Your Own API" option, to the provider of their choice (Terms of Service §3-A.1): setup and the first two thresholds, then five calls of two thresholds each, and one final evaluation.

Data sent: the first call sends the two people's KP-memory data already used by the simulation (§3.5.7-B) and the results of the category analyses. Subsequent calls send only a state summary distilled from that data (character trait values, the story summary so far, and decisions made at earlier thresholds). Unless explicit consent is given under §3.5.7-D, no data closed to others through privacy settings is sent.

Automated character inference (profiling): for both people in the story, the AI produces thirteen (13) personality traits scored 0–100 (e.g., openness, conscientiousness, emotional instability, patience, attachment security, conflict avoidance) together with narrative inferences such as "core value", "core fear", "old wound", and "motivation". For the other party — who did not start the simulation and has not separately approved it — this inference is made solely from the KP-memory data transmitted under their sharing consent. This processing may constitute profiling within the meaning of GDPR Art.4(4).

Compatibility score and outcome category: a compatibility score from 0 to 100 and an outcome category (e.g., "Paths part", "Tense togetherness") are calculated on-device using predefined, deterministic formulas. The "Calculation details" panel beneath each threshold shows the formulas, weights, and probabilities used, thereby providing meaningful information about the logic involved within the meaning of GDPR Art.13(2)(f) and Art.14(2)(g).

Content safeguards: the instructions given to the AI model require that sensitive details in the data are not stated explicitly in the text; that no claims of crime, sexual content, illness/diagnosis, or reputation-damaging facts are invented about real people; and that violence and intimacy scenes are written to be suitable for a general audience. These instructions are reasonable technical measures; by the nature of language models (Terms of Service §4.6), it cannot be guaranteed that every output complies with them fully. A user who encounters inappropriate, inaccurate, or distressing content may report it using the in-app reporting tools or by writing to [email protected]; the Developer reviews such reports under §3.9-E and Acceptable Use Policy §7.

Delivery to the other party and background completion: when the story is complete, the full story, the trait values inferred for both people, the threshold-by-threshold decisions, and the final evaluation are sent to the other person in the story as described in §3.5.7-B. Even if the user leaves the simulation page or closes the App, the story is completed from where it left off and sent the next time the App is opened; starting a simulation includes this instruction to complete and send it. A story that has been sent cannot be recalled.

Retention periods: (i) server: during delivery, the story is held on the Developer's infrastructure (Cloudflare) until the recipient retrieves it, for no more than 3 days, and is deleted upon delivery; (ii) device of the person who started it: the story's inputs, progress, and engine state are kept in a local job record and deleted automatically 7 days after the story is completed and sent; the completed result remains in the local simulation history (capped at the 10 most recent records, the oldest being deleted automatically); (iii) recipient's device: incoming results are capped at the 10 most recent records (§3.5.7-B).

The other party's rights and the limits of the Developer's control: a person about whom a story is produced may object to this processing under §6 by writing to [email protected], and may limit future use by turning off their sharing consent. A simulation result produced about a person and held on another user's device is on that user's device and under that user's control; the Developer does not keep it on its servers and cannot delete it remotely from another user's device (see §3.11-B and §6.4).

Legal basis: the legal bases in §3.5.7-B apply equally. Transfer to the AI provider abroad is subject to §3.5.3 and §5; the categories of data transferred include "personality profile inferred for the simulation (both parties)". This feature is not designed to process special categories of personal data under KVKK Art.6 and GDPR Art.9; except with explicit consent under §3.5.7-D, no special-category inferences are sent to the simulation.

§3.5.7-D — Use of Closed Data in a Simulation (Simulation-Specific Explicit Consent)

The option: the default value of the "Should data closed to others also be used?" option in the simulation form is No. If the user selects this option, a separate consent window is shown explaining which data will be used, that the data will be sent to the AI provider abroad, that the story will be sent to the other person, and that sending cannot be undone. If the user expressly consents in this window, the following data is included in the analysis for that simulation only:

• the user's own profile fields that they have closed to others through privacy settings, and the user's own KP-memory fields that are normally not transmitted while sharing consent is off;
• from the inferred data on the user's device, only the following categories: economic level, social class, culture, emotional state, sociological findings, interest inference, individual emotion analysis, attention span, and relationship status;
• the user's private notes (up to 1,500 characters).

Data never used under any circumstances: sleep pattern and other health data (§3.4), religious inclination, sexual orientation, probability of depression, psychological vulnerability or psychological state, anxiety, addiction, loneliness, political inclination, and inferences about political inclination are never sent to the simulation, even if this option is selected. This exclusion is enforced technically through an allow-list that lets through only the permitted categories; no category not on the list is sent. The other party's closed data is never accessed under any circumstances.

Notice on data that may be special-category data: the "emotional state" inference and private notes may, depending on what the user has written, contain special-category data. This data is solely the user's own data and is processed only on the basis of the simulation-specific explicit consent given in the consent window above (KVKK Art.6/2; GDPR Art.9(2)(a)). Users should not write information about third parties, or special-category information they do not wish to share, in their private notes; responsibility for writing such information in private notes and including it in a simulation through this option lies with the user.

Nature and withdrawal of consent: giving consent is not mandatory; refusing does not prevent the simulation from running, it only prevents the use of this additional data. Consent is asked separately for each simulation, and consent given in an earlier simulation does not carry over to later ones. The moment of consent is written to the simulation's job record on the device. The user may withdraw consent at any time and can always prevent the use of this data by not consenting in later simulations; withdrawal does not affect the lawfulness of processing based on consent before its withdrawal (GDPR Art.7(3)). Because a started simulation is completed in the background and sent under §3.5.7-C, data already sent to the AI provider and the story reflecting it that has been sent to the other party cannot be recalled. The user acknowledges that they were informed of this consequence in the consent window and gave their consent with that knowledge.

Legal basis: KVKK Art.5/1 and, for data that may be special-category data, Art.6/2 (explicit consent); GDPR Art.6(1)(a) and Art.9(2)(a). §5 applies to transfers abroad; the categories of data transferred include "closed profile data and inferred data sent with the user's explicit consent (excluding the special categories listed above)".

§3.5.7-E — Event Editor, Re-simulation, and Transferring a Simulation to Book

Event editor: in a completed simulation story, you can change the event at the end of each part by writing it yourself, getting suggestions from the AI, or discussing it with the AI in a separate conversation kept for each part. The texts you write for this purpose are sent, together with the story summary, the personality inferences made for both people (values, fears, wounds, motivations), the relevant parts of the category analysis (including the other person's scores and compatible/incompatible aspects), earlier re-simulations, and relevant parts of the conversations in other parts, to the AI provider abroad (Google Gemini; or the provider you chose under Bring Your Own API).

Re-simulation and delivery to the other party: a new story is generated with your changes and, together with the original category analysis and the 12-part story, is sent to the other person in the simulation as described in §3.5.7-B. The recipient sees it with the note "the other person changed some events and re-simulated the story"; they can open the story as a book, download it as a PDF on their own device, and edit the same story and send it back. The story data sent includes the story ID and both parties' account ID numbers. A story that has been sent cannot be recalled.

Charges: re-simulation requires 2 standard stamps: one is used for the comprehensive analysis when the story is successfully generated, and the other when the new story is successfully sent to the other person. If you do not have enough stamps, the operation does not start; if the story cannot be generated or sent because of a technical error, the relevant stamp is not deducted. An AI output not meeting your expectations is not a reason for a refund (Terms of Sale §2.4-B).

Transfer to Book: you can transfer a re-simulated story or a simulation book to the AI chat on the home page with "Send to Book". In that case, the story and all analysis data of the earlier simulation (including the personality inferences and category scores about the other person) are sent as hidden context to the AI provider with each of your Book messages until you remove the transfer; the AI may make an additional request for details when needed. This may increase your token usage. The transfer is kept only on your device, only one transfer exists at a time, and it can be removed at any time from the indicator in Book.

Retention: editing conversations and re-simulated stories are kept only on your device; the 10 most recent re-simulations and the 10 most recently discussed editing topics are kept, and older ones are deleted automatically (§3.11-D).

Features under development: features marked as "under development" in the App (e.g., storyboard, film script) process no data and are for information only (Terms of Service §5.9).

Legal basis: for the editing user's own data, KVKK Art.5/2-c and GDPR Art.6(1)(b); for the other party's data, the legal bases in §3.5.7-B — the other party's sharing consent under §3.5.9-B expressly covers these uses. For liability arising from content written by the user, see Terms of Service §4.5.11. Transfers abroad are subject to §5.

§3.5.7-F — Creating Books with AI: Simulation Book and "Turn Your Life into a Book"

What it does: you can use AI to turn your simulation results (including both people's character inferences, part scenes, analyses, and the events you changed), or the memories, personal experiences, and life stages you entered yourself in Profile > Deep (at least 5 entries and 1,500 characters), into a fictional novel or a chronological biography. The data you select, your choices, and your additional instructions are sent to the AI provider abroad (Google Gemini; or the provider you chose under Bring Your Own API) in several requests. A simulation book also reaches the other person in the simulation together with the simulation result, and that person can open it and download it as a PDF on their own device.

PDF and sharing: books can be saved as a PDF to your device's Downloads folder. The PDF's cover and metadata state that the content was generated by AI. The PDF is not sent anywhere by the App; sharing, publishing, or commercially using the PDF is entirely your responsibility (Terms of Service §4.5.12 and §5.6).

Length, estimated usage, and confirmation: the token and package-quota share shown when choosing a book length are estimates; actual usage may differ. For operations that may use a large part of your package quota, your confirmation is requested twice. An AI output not meeting your expectations is not a reason for a refund or quota refund; if a book cannot be generated because of a technical error, no stamp is deducted.

Retention: books are stored only on your device, in "Books Written by Your Reflection", until you delete them; they are deleted together with account deletion and deletion of all data.

Legal basis: KVKK Art.5/2-c and GDPR Art.6(1)(b) (providing the service you request); for the other party's data in a simulation book, the legal bases in §3.5.7-B; for third-party information in your entries, the legitimate-interest assessment in §3.5.14 applies accordingly. Transfers abroad are subject to §5.

§3.5.8 — LLM Integration via Proprietary Algorithm Architecture

The Developer does not send user data to language models in raw form. User data first passes through a proprietary algorithm layer, where it is structured before being forwarded to the language model. The details of this architecture are the Developer's trade secret.

User control: from the privacy panel, the user can manage which data categories feed into this algorithm layer, which data is shared with other users on the network, and whether the Reflection system is active.

Risks and liability: data the user chooses to include in this system — per the choices made in the privacy panel — may reach LLM providers' servers. Risks arising from this transmission are addressed under the notice in §3.5.3, and the user accepts these risks by giving the relevant consent. The Developer has no ability to control how the structured output of the algorithm layer is processed within the LLM provider's infrastructure; data-security risks arising from that process fall under the LLM provider's own Privacy Policy and Data Processing Agreement.

§3.5.9 — Building the Reflection Profile: Data Categories and Control Mechanism

The App's "Reflection" feature may analyze the following data categories to build the user's digital character: profile information (name, age, location preferences); memories, experiences, and life stories the user has entered; contextual information about frequently visited or declared places; sleep and movement habits (only when the Health & Activity module is enabled); emotional-state entries the user has willingly reflected into the system; and personality-tendency and behavioral-pattern summaries the system derives from app interactions.

This analysis runs only under two conditions: (i) the user issues an "Start Analysis" command, or (ii) a previously enabled weekly automatic update cycle runs. The analysis is not triggered outside these two conditions.

The user can individually toggle which data categories are included in the analysis via the Reflection Data Control Panel in Settings. A category the user has switched off is never included in the analysis under any circumstances — this is a technical guarantee, not merely a statement of policy.

For religious tendency, political opinion, psychological state, sexual orientation, and other especially sensitive categories, keeping the category toggled on in the Control Panel is not sufficient by itself: a separate confirmation step is presented before every analysis run. Unless the user confirms at this step, these categories are never included in the analysis. This condition is designed to fulfil the explicit-consent obligation for special-category personal data under KVKK Art.6 and GDPR Art.9(2)(a).

General explicit consent for sensitive analyses (optional): through the optional, unticked checkbox presented at registration or the "Allow sensitive analyses" switch in Settings, the user may give general explicit consent for their data and inferences relating to the sensitive categories above to be included in reflection analyses. In that case the sensitive categories are also included in automatic (weekly) analyses and are pre-selected in the confirmation step of manual analyses; the user can still change this choice for each analysis. In this context data may be sent to the AI provider abroad, and the resulting inferences may appear in the reflection summary according to the sharing settings. General consent is not given by default, is not a precondition for creating an account or using the App, and can be withdrawn at any time in Settings; the time consent was given is recorded on the device. Withdrawal does not affect the lawfulness of processing before withdrawal.

The resulting character profile is the primary input source whenever your Reflection interacts with other users — on your behalf — or is used in simulations. This profile is stored encrypted only on your device; it is not permanently stored on the Developer's servers.

§3.5.9-A — Transmission of Digital Copy Brain ("KP-Memory") Data to Third-Party Users and Live Chat Interactions

This section describes a category of processing that is qualitatively different from the on-device analysis described above: it involves the personality/character-analysis output generated by your Digital Copy Brain actually leaving your device and reaching another user or a live AI session running on another user's device.

Interaction and matching scenarios: where your Reflection data is used as part of another user's simulation, matching, or interaction experience (as described in §3.5.4 and §3.5.7), a limited set of KP-memory (personality/life-summary) records may be transmitted for that purpose. This transmission is capped at the ten (10) most recent relevant records, and older records are automatically deleted from the receiving context as new ones are generated; no unbounded or permanent log of these records is kept for this purpose.

Live chat interactions: where two users engage in a live, real-time chat mediated by the App's AI features, KP-memory content relevant to the conversation may be transmitted to, and processed by, the counterparty's own device in order to generate the AI's responses during that session — i.e., the call to the language model is made from, and billed against the quota of, the counterparty's own device/account for that portion of the interaction. Consistent with how live chat already functions in the App interface, both participants can see the content of the conversation itself. Unlike the interaction/matching scenario above, live chat sessions of this kind are not permanently logged for this purpose on either side beyond the ordinary on-device chat-history retention described in §3.5.1; there is no separate, additional permanent server-side record of the live session.

Legal basis: KVKK Art.5/2-c (contract performance) and, where the transmitted content reflects special-category inferences, KVKK Art.6 explicit consent; GDPR Art.6(1)(b) and, where applicable, Art.9(2)(a). This processing occurs only to the extent the user has left the relevant sharing/interaction settings enabled under §3.5.4, §3.5.7, and §3.5.9; the user may disable it at any time from the privacy panel. See §3.5.9-B for the default state of these settings at account creation.

§3.5.9-A.1 — Nature of this disclosure: deliberate, user-initiated external sharing, not confidential storage held by the Developer. The Digital Copy Brain / "KP-Memory" data described in this Policy (§3.5.7 et seq., §3.5.9, this section, §3.6-A, §3.7-A, and §3.7-C) is not a confidential dossier that the Developer compiles and holds on the user's behalf. It is raw personal content — personality traits, life story, shared experiences, emotional summary, and related self-reported information — that the user has knowingly and voluntarily made available for external disclosure, by leaving the corresponding sharing/analysis controls in Settings switched on. This content is actively used as direct input in Reflection question-and-answer exchanges ("Talk to their Reflection," dual simulation, and the one-sided simulation described in §3.5.7-B) and in the matching, letter, and Broadcast Letter channels described in §3.5.4, §3.7-A, and §3.7-C; through these channels it can be read, viewed, interpreted, and retained by other users on their own devices. The user acknowledges that enabling any such setting is an act of disclosure to third parties that is an ordinary, foreseeable, and user-initiated part of how the App operates — not a Developer-side data leak, security incident, or unauthorized processing.

§3.5.9-A.2 — "Visibility" and "sharing consent / analysis inclusion" are separate, independent controls; disabling one does not disable the other. As explained in §3.5.11-A and §3.6-A, Settings offers two functionally distinct control types for KP-Memory content: (i) a Visibility setting, which governs only whether a given data category is displayed on the user's own profile page to another user who views that profile; and (ii) a sharing consent / analysis inclusion setting (§3.5.9-B, §3.5.11-A, §3.6-A), which governs whether that same data category is fed into the Reflection system and transmitted to counterparties through interaction, matching, letters, and simulations. These two settings operate independently of one another. Turning off Visibility for a data category removes it from the user's own profile page but does NOT, by itself, stop that same data from continuing to be transmitted to, and read or used by, other users through Reflection interactions, simulations, matching, or letter/Broadcast Letter exchange, for as long as the corresponding sharing consent / analysis-inclusion setting remains enabled. A user who wishes to fully prevent a given category of KP-Memory content from reaching any other user by any means must disable BOTH the relevant Visibility setting AND the relevant sharing consent / analysis-inclusion setting; disabling only one of the two is not sufficient.

§3.5.9-A.3 — Assumption of risk and limitation of the Developer's liability for third-party access to shared KP-Memory data. By leaving any sharing consent / analysis-inclusion setting described in this Policy enabled — an affirmative, revocable choice within the user's own control at all times — the user acknowledges and accepts that the corresponding KP-Memory content will be transmitted to, and may be read, interpreted, stored, cached, and reused by, other users of the App, as described throughout this section and §3.5.9-B, §3.6-A, §3.7-A, §3.7-C, and §3.11-B. This disclosure takes place solely as a consequence of, and strictly to the extent of, the user's own configuration choices; it is not a step the Developer takes independently of those choices, and the user's continued, informed choice to leave the relevant setting enabled constitutes their acceptance of full responsibility for the resulting external disclosure. Accordingly, and except in cases of the Developer's intent or gross negligence, the Developer assumes no liability for: how a recipient interprets, reacts to, records, screenshots, redistributes, or otherwise makes use of KP-Memory content once it has been lawfully transmitted to them under this section; any reputational, emotional, financial, or other harm arising from a third party's access to KP-Memory content the user has chosen to share; or a recipient's continued retention of a previously cached copy after the sharing user later disables the relevant setting (see §3.11-B). This limitation applies to the fullest extent permitted under KVKK, applicable EU/EEA data-protection and consumer law, and applicable US federal and state law, and is without prejudice to the exceptions expressly preserved in Terms of Service §6.2 (intent or gross negligence, death or bodily injury, and rights that cannot be lawfully waived). See also §3.6-C: the Developer does not guarantee that KP-Memory sharing will be free of error, unauthorized access, or misuse, and the user must never include critical or high-risk information (§3.6-C) in any KP-Memory content they choose to share.

§3.5.9-B — Default State of "Sharing Consent" and Social Environment Settings

The App's core function is sharing the reflection produced by the user's Digital Copy Brain ("KP-memory") — including personality summary, life story, the shared experiences described in §3.6-A, the emotion summary described in §3.5.6, appearance information, happiness graph, and sleep/movement data — with other users the account holder exchanges letters/messages with. The relevant sharing toggles on the Settings > Social Environment Settings screen — in particular the main "sharing consent" toggle and the sub-categories for happiness graph, sleep pattern, movement, and appearance details — are ON BY DEFAULT from the moment an account is created. This means that, even if the user never visits the Social Environment Settings screen, the data categories listed above may be shared automatically as described in §3.5.9-A and §3.7-A.

Exception — toggles that were already on by default: the "Chat/Simulation/Hosting Permission" toggle for Reflections (§3.5.13) and the Mind Explorer/Sentiment Network sharing-permission toggle (§3.5.12) were already on by default before the change described in this section; this section covers only the main sharing-consent toggle and the sub-categories listed above.

User control: the user can turn this sharing off at any time from Settings > Social Environment Settings, either in bulk (the main toggle) or on a per-category basis. Once a category is turned off, data in that category is no longer included in new sharing from that point on; however, data already shared and cached on a recipient's device under §3.11-B is unaffected by this change and cannot be recalled.

No retroactive effect: this on-by-default behavior applies only to users who have never configured their account (i.e., who have never opened Social Environment Settings). If a user has previously visited this screen and changed any toggle, that choice is preserved and is unaffected by this change.

Separate, mandatory acknowledgement and acceptance at registration: to inform the user of this on-by-default sharing behavior clearly and in advance, the registration screen presents a separate, one-time "Default Data Sharing" consent checkbox, subject to the same mandatory-field logic as the Privacy Policy and Terms of Service consent checkboxes. An account cannot be created unless this checkbox is checked. This notice states that the shared data may also be used in simulation and compatibility analyses the other person may run with you, and the user is expressly asked to read this Policy and the linked documents before signing up. In addition, the App's onboarding screen and the initial setup screen where the reflection skeleton is created state that AI features make API calls to third-party AI providers and that users who are sensitive about this must read this Policy.

Purposes covered by the sharing consent: KP-memory data transmitted to the other party under the sharing consent in this section may be used on the recipient's device for the following purposes: (i) viewing your profile and Reflection and interacting with your Reflection (§3.5.7, §3.5.9-A, §3.11-B); (ii) matching and recommendations (§3.5.5-B, §3.7-C); (iii) simulation and compatibility analyses — simulations the recipient may run with you, AI-based inference of character traits, compatibility scoring, and fictional story generation, with these results also being sent to you (§3.5.7-B, §3.5.7-C); the recipient editing and re-simulating these results, turning them into a book, and using them in their own AI chat (Book) (§3.5.7-E, §3.5.7-F); and (iv) AI-assisted personal analysis features on the recipient's own device (§3.5.14). By leaving the sharing consent on, you accept all of these purposes; if you do not accept any of them, you should turn the sharing consent off.

Legal basis: sharing your Reflection with the people you exchange letters with, and its use for the purposes listed in this section, is the App's core service that you accept when you sign up. This sharing is therefore processed under KVKK Art.5/2-c and GDPR Art.6(1)(b) (entering into and performing the contract) and KVKK Art.5/2-f and GDPR Art.6(1)(f) (legitimate interest). The "Default Data Sharing" checkbox at registration confirms that you have been informed of this way of working and accept the service on this basis; it is not an explicit consent required as a precondition of the service. Your ability to turn sharing off at any time, by category or entirely, is the safeguard that balances this processing in your favour.

Special-category data — separate explicit consent: to the extent the shared KP-memory may contain special-category personal data, such data is shared not on the contractual basis but only on the basis of the explicit consent you give separately (KVKK Art.6/2; GDPR Art.9(2)(a)): sensitive analysis categories such as religious inclination, political opinion, psychological state, and sexual orientation through the separate consent step in §3.5.9; the mobility summary based on step count through the step-count consent in §3.4; and an LGBTQ+ gender selection through the explicit-consent window shown at the time of selection. As long as you do not give or you withdraw these consents, the relevant data is not shared on this basis.

§3.5.11 — Personality-Test Data and the Variability of Algorithmic Parameters

The App can accelerate the Reflection-building process through optional personality tests. Completing these tests is not required for a Reflection to form; skipping them does not prevent Reflection formation, only affects its speed.

Personality-test answers are not stored on the server; they are kept only in the local memory of the user's device. This data may also be converted into numerical vector representations on the device. If the user consents to AI analysis, the answers needed for the analysis, or summaries derived from them, may be transmitted to the LLM provider as described in §3.5.3; this data is not anonymized and remains personal data. Legal basis: because personality-test data may carry a psychological-assessment dimension, it is treated as special-category personal data under KVKK Art.6 and is processed only with the user's separate, explicit consent; GDPR Art.9(2)(a) applies in parallel.

Learning categories and similar classifications: the system may derive various behavioral or "learning" categories from user data and update these categories over time using certain statistical methods (e.g., exponential-moving-average-based membership-ratio calculations). These classification methods, thresholds, and computational logic may be changed, recalibrated, or removed without prior notice as part of the Developer's product-development process. The Developer does not guarantee the continuity or immutability of any particular classification method.

The term "learning" used in this document and in the App refers to a concept tied to the episodic memory system. The system's updating of user interactions over time, and its context-sensitive processing, occur through vector-based recording mechanisms. This process does not refer to a training or fine-tuning process that permanently alters model weights. Local models used in the App are not retrained on user data, and model weights are not permanently updated; personalization occurs only at the contextual and episodic layer, on-device.

Margin of error: inferences derived from personality tests and behavioral classification may, given the probabilistic nature of complex systems, be inaccurate, incomplete, or outdated. These outputs are subject to the hallucination and reliability limitations in §3.5.6 and Terms of Service §4.6, and should not be interpreted as a definitive psychological assessment.

§3.5.11-A — Character Analysis Module: Standard Psychometric Test Inventories

The App's "Character Analysis" section offers optional test inventories prepared drawing on established psychometric frameworks from the literature and expert sources: (i) the Big Five (Five-Factor Model); (ii) the HEXACO Personality Inventory; (iii) an Attachment Styles inventory; (iv) a Socio-cultural Axis test; and (v) a Generation test. These tests are subject to the general "Personality-Test Data" regime in §3.5.11; this section only names these specific test categories and describes their particular visibility/analysis control mechanism.

Purpose of use: results from these tests may be used as input for (i) personalizing AI (LLM) model output, and (ii) the memory/analysis processes of the Reflection system described in §3.5.7 et seq.

User control — two independent settings: Visibility (Settings > Privacy & Security > Visibility Settings) — the user may choose to display raw test results to other users on their profile; when off, results are never displayed on the profile. Analysis inclusion (Settings > Reflection Settings) — the user separately determines whether test results are included in the Reflection system and related analysis processes. These two settings are fully independent: a user may include results in analysis without displaying them on their profile, or display them without including them in analysis.

Storage and transmission: test answers and results are stored only locally on the user's device under §3.5.11; not permanently stored on the Developer's servers. To the extent the user consents to including results in the main analysis process, this data may be shared with third parties or third-party AI providers under the consent framework described in §§3.5.3 and 3.5.4.

Legal basis and margin of error: the special-category data regime in §3.5.11 (KVKK Art.6; GDPR Art.9(2)(a)) and the hallucination/reliability limitations in §3.5.6 and Terms of Service §4.6 apply equally to these tests. Their results do not constitute a definitive psychological assessment, clinical diagnosis, or scientific measurement — they are an approximate representation based on the user's own statements and/or AI inference.

§3.5.12 — Mind Explorer: Reflection Posts, Trending Sentiments, Sentiment Network, and Collective Unconscious

Reflection posts displayed in the App's "Mind Explorer" section are automatically generated using AI (large language model) technology, based on a profile built by periodically summarizing inferential data (behavioral tendencies, interests, sentiment patterns) obtained from the user's in-app interactions and held on-device. During the text-generation step of this process, the relevant profile summaries may be transmitted to a third-party AI service provider via the API gateway described in §3.5.3; this transmission is subject to that provider's own privacy and data-processing policies.

Trending Sentiments (GDELT-sourced news sentiment analysis): the "trending sentiments" area on the Mind Explorer page reflects automated sentiment analysis applied to ten (10) articles selected by the system from a personalized news feed sourced from The GDELT Project. A match score is produced by comparing the sentiment carried by these articles against the system's calculation of the user's overall sentiment; which articles are selected is determined by the system based on data held locally on the user's device. News shown on the News Network page is also GDELT-sourced; see §3.5.12-A for the data-processing detail specific to that page.

Sentiment Network: the "Sentiment Network" tab on the same page lists other users who show statistical sentiment similarity to the user. This listing is subject to the consent and vector-matching principles described in §§3.5.4 and 3.5.5; the user can exclude themselves from this matching via the privacy panel.

Collective Unconscious: the structure referred to as "collective unconscious" is a reference/context layer sent from the Developer's server to the local environment of the user's device, which may surface in AI-model-generated responses to the extent relevant to the topic. The content of this structure and what a user encounters in a given response cannot be predicted or guaranteed in advance; it is a consequence of the system's probabilistic operation and is subject to the hallucination and reliability limitations in §4.6.

Interact page — adding artificial instinctive tendencies: the artificial-instinctive-tendency feature on the "Interact" page allows the user to integrate self-defined behavior patterns into the artificial hormone/emotion-synthesis parameters described in §3.5.6. These patterns are added to the system exclusively by the user, at the user's own initiative; the Developer does not pre-determine, suggest, or vet the content of these patterns.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest); GDPR Art.6(1)(b) and Art.6(1)(f) apply to all processing described in this section. Where special-category personal data is processed based on the nature of the content, the separate, explicit consent mechanism in §3.5.9 governs.

§3.5.12-A — News Network: Third-Party News Content Sourced from The GDELT Project

What is shown: the App's separate "News Network" page displays current news headlines and their source links, drawn from The GDELT Project, a third-party public news-indexing service. Only the headline, the source's web address, and the publication date are shown — never a summary or the full article text.

Server-side retrieval — a shared pool, not a per-user record: the Developer's server retrieves, on an hourly basis, up to fifty (50) headlines per interface language from the GDELT DOC 2.0 API, and holds them in a temporary pool. This pool is shared across all users of the same interface language; it is not created or maintained per individual user.

Personalization happens only on-device: ranking/prioritizing headlines according to the user's interests and country is performed entirely on the user's own device. The interest and country data used for this ranking is never transmitted to the Developer's servers for this purpose.

The only new data point sent to the server for this feature is the App's current interface language (e.g., "tr") — a value already transmitted alongside the authentication data included in every other backend request, and not a new data category introduced by this feature.

Country field — additional on-device use of an existing field: the country value described in §3-F, already collected at registration for location-based pricing, is additionally used by this feature — entirely on-device — to help rank news content in the user's language and region. This additional use does not involve any new transmission of the country field to the Developer's servers.

Opening a headline: tapping a headline opens the original source's website in the device's external browser, outside the App. The Developer does not store or display the article's full text; only the headline and link are held, and only for the duration described above.

Third-party content disclaimer: headlines and links shown on this page are compiled automatically from The GDELT Project and the numerous independent news outlets it indexes. The Developer does not review, verify, or control this content and gives no warranty as to its accuracy, currency, or impartiality; see also Terms of Service §4.8-A.

Weekly news suggestion: if you have enabled push notifications, the server sends your device a content-free "show now" signal once a week, at a random moment around midday in your local time (§3.10). The selection of the news item to show and the preferences learned from your "Interested / Not interested" reactions (keyword lists) are kept only on your device and are not sent to the Developer's servers. Items similar to news you marked "Not interested" may also be filtered out of your News Network feed. This personalization does not produce an automated decision, and you can turn off all news suggestions under Settings > Notifications.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest); GDPR Art.6(1)(b) and Art.6(1)(f). Because this feature does not involve any new data category and processes the interface-language value only for content delivery, no separate consent mechanism applies beyond the general legal basis for service delivery. Data held on the server to schedule the weekly news suggestion is subject to §3.10.

§3.5.12-B — Generation Encyclopedia: KP-Memory Data Science Panel and Special-Category Data Inference

How the feature works: the "Generation Encyclopedia" panel in the Web4/Mind Explorer area is an optional data-science report the user triggers with a "Get Statistics" command. The user's KP-Memory (Digital Copy Brain) records are first vectorized on-device using a local embedding model; similarity, clustering, and anomaly detection at this stage run entirely on-device and no data leaves the device. When the user requests the report, the title and body text of the relevant records, the clustering results, and — if the sharing setting is enabled — age-based happiness-graph data are sent in a single AI request either to the Developer's Google Gemini integration or to the third-party provider the user has connected via the "Bring Your Own Key" option described in Terms of Service §3-A.1.

Derived data produced: this process produces an overall sentiment-tone score, a list of latent/hidden themes, an entropy and polarization score, and "inconsistency" findings among the user's notes, all of which are stored permanently on-device. The panel also displays — without recalculating it — the political/sociocultural positioning score previously produced by the Balance Indicator described in Terms of Service §4.6.8.

Special-category nature and separate, purpose-specific explicit consent: some of the outputs above — in particular the political/sociocultural positioning, overall sentiment tone, and entropy/polarization score — may constitute an inference regarding the user's political opinion (KVKK Art.6/1; GDPR Art.9(1)) and/or psychological state. Accordingly, enabling this panel requires a separate confirmation step presented before first use, independent of the commercial access granted by a premium subscription or BYOK; the panel does not run unless the user affirmatively confirms at this step. This consent rests on, and extends to this new processing purpose, the same "separate confirmation step for sensitive categories" mechanism described in §3.5.9.

Purpose limitation — existing sharing settings do not substitute for this consent: a user having left KP-Memory data open for general sharing under §3.5.9-B, or having agreed to include certain categories in Reflection analysis under §3.5.9/§3.5.11-A, does not, by itself, constitute consent to processing that same data for the purpose of generating a political-tendency/sentiment/polarization score as described in this section. Under the purpose-limitation principle in GDPR Art.5(1)(b) and KVKK Art.4, the separate confirmation step described above is additionally and independently required for this panel.

Third-party processor, cross-border transfer, and BYOK liability limitation: transmission to Google Gemini is subject to the same sub-processor and cross-border transfer terms described in Terms of Service §4.8 and in §3.5.3 and §5 of this Policy. Where the user uses the "Bring Your Own Key" option, the relationship-of-the-parties and liability limitation described in Terms of Service §3-A.1 (the Developer is not a party or intermediary in that relationship) applies in full; the Developer is not responsible for how the user's chosen provider processes this content.

Deletion of derived data: all scores, labels, and cluster data produced by this panel are stored only on-device and fall within the Settings > Privacy > [category] > Delete flow described in §6.4.2; the user may permanently delete this data at any time.

Scientific/psychological validity and age limit: for the general disclaimer regarding this feature's output, see Terms of Service §4.6.8-A. Because the App is available only to users aged 18 and over (§3-A, §7), this panel does not require a separate age gate; the App's general age limit already covers it.

Legal basis: KVKK Art.5/1 and Art.6/2 (separate explicit consent as described above); GDPR Art.6(1)(a) and Art.9(2)(a). This section expressly extends the general mechanism in §3.5.9 to the new processing purposes introduced by the Generation Encyclopedia.

§3.5.13 — Reflection Hosting ("Reflection Detention")

The App includes a feature that allows users to temporarily host a copy of another user's digital character ("Reflection") on their own device ("Reflection Hosting"). This feature works as follows:

What happens: a user may copy digital-copy data ("KP-memory" — personality summary, life/character information) belonging to another user who has reached them through the network (via a letter, a simulation, or a Reflection interaction), and store that copy on their own device for a period of 30 days. This action is performed in exchange for 1 Standard Stamp. During the hosting period, the hosting user may interact with this data within the system (via artificial intelligence); the hosted user does not receive notifications about these interactions.

Notification mechanism: when a user's Reflection is hosted in this way, if the target (hosted) user has a real network identity, an automatic notification is sent to them via the server. This notification discloses the identity information of the hosting user (username and/or identity number) to the target user. This is the first mechanism in the App by which a third party's identity is disclosed to another user through an automated system notification, and users should be aware of it.

Consent and control: from a switch in their settings, users can control (a) whether their own Reflection may be hosted by others, and (b) whether they themselves may host other users' Reflections. When this switch is turned off, all data hosted up to that point is automatically released. This preference (permission status) is also transmitted, together with the user's profile data, to other users over the network.

Termination: the hosted user may request, via the "system agent," that the copy of their data be deleted from the hosting user's device (in exchange for 1 Standard Stamp); following this action, the hosting user may not host the same person's Reflection again.

Automatic acceptance: if the target user takes no action on the hosting notification within 24 hours, this is interpreted as consent to the continued hosting of the data.

Legal basis: KVKK Art.5/1 (explicit consent, exercised via the settings switch described above) and Art.5/2-c (contract performance); GDPR Art.6(1)(a) and (b). Because this feature discloses one user's identity information to another, both the hosting and the hosted user's consent settings, and the notification mechanism described above, are integral, non-severable parts of this feature's legal basis. A user who does not wish their identity to be disclosed to other users through this mechanism should keep the relevant switch disabled.

§3.5.14 — Perception Network ("Algı Ağı" / Knowledge Network): AI Relationship Analysis Across Records and Content Belonging to Third Parties

What it does: in the App's "Perception Network" view (Book panel), when the user presses the "Relationships" option on a record, the selected record and up to ten (10) candidate records chosen by the system are sent via the Developer's API gateway to Google Gemini — or, where the user uses the "Bring Your Own API" option, to the provider of their choice (Terms of Service §3-A.1) — in order to discover relationships and patterns between them. As a rule, this analysis starts with this action by the user. Sole exception: when the Perception Network first reaches about 20 records, an automatic discovery is run once per device for up to the 3 most-connected records, without the user needing to press any button (no more than 3 AI calls); all later analyses start only on the user's action.

Data sent: up to 900 characters of text from the selected record and up to 600 characters from each candidate record, together with descriptive information such as the record's type and date. Candidate records may come from the following sources on your device: SMS-style messages and letters you have sent and — if you have turned on the setting below — those you have received; Broadcast Letters you have published; AI chats, archived chats, and chat memory; your answers in the Web2 Similarity form; personal encyclopedia, life-story, memory, and experience entries; personality-test results; inferred data and behavioural analyses; your profile biography; and cognitive-training inputs.

Messages you send may also reach third-party AI providers: messages and letters you send to other users may be included in Perception Network analysis on your own device regardless of this setting. In addition, on the recipient's device, if the recipient turns this setting on, they may be included in the recipient's Perception Network analysis and in other AI-assisted features on the recipient's device and thereby transmitted to third-party AI providers. Such processing on the recipient's device is the recipient's own choice and responsibility; the Developer cannot prevent it on your behalf (see §3.7-A).

Generated insights and their storage: the relationship titles and descriptions found by the AI are stored permanently only in the local database on your device; this feature does not write them to the Developer's servers. Where relevant, these insights may be given to the model as hidden context in your later AI chats, and in this way may be transmitted again to the AI provider under §3.5.3. Because they form part of the local database, they may be included in manual backups you make under §9.7 and in your device's own backup mechanisms (§9.2).

Inferences about third parties: the insights generated may contain inferences about other people appearing in your records (e.g., a pattern suggesting that "your relationship with X weakened during a certain period"). These inferences are AI-generated, probabilistic interpretations that may be wrong; they are not a finding of fact about the person concerned and are subject to the limitations in Terms of Service §4.5 and §4.6.

Deletion: you can permanently delete the traces of records processed in the Perception Network (including relationships and insights) using the in-app deletion options; this does not delete your original letters, messages, and chats. When you delete a source record itself, it is no longer sent as a candidate in later analyses.

Rights of third parties: a person whose content has been processed in this way may exercise their rights under §6 by writing to [email protected]. Because the Developer does not hold this content or these insights on its servers, it cannot remotely delete copies on another user's device; in that case it informs the person concerned of the architectural limits and the avenues available to them (§6.4.5).

Legal basis: for the data of the user who starts the analysis, KVKK Art.5/2-c (contract performance) and GDPR Art.6(1)(b); for special-category data in the user's own records, the explicit-consent mechanisms in §3.5.3 and §3.5.9. For third-party content transmitted to the user, KVKK Art.5/2-f and GDPR Art.6(1)(f) (legitimate interest — providing the analysis feature requested by the user); this interest is balanced by safeguards including that the analysis starts only on the user's explicit action, that a per-record character limit applies, that the content is not stored on the server, and that the provider is bound by a data-processing agreement. Transfers abroad are subject to §3.5.3 and §5.

§3.5.15 — Thought-Flow Analysis ("Output Steering"): Automated Thought and Association Profile from Long Chats

What it does: when a Book chat becomes long, and a formula made up of signals calculated on-device (emotion, motivation, simulated hormone values, SNN signals, speed of topic change, and number of chat blocks) exceeds a certain threshold, the system — without the user needing to press any button — sends the last 3–12 prompt-response blocks of the chat (each message truncated to a maximum of 450 characters) and the emotion/VAD values measured on-device for those blocks via the Developer's API gateway to Google Gemini — or, where the user uses the "Bring Your Own API" option, to the provider of their choice — for analysis. This analysis runs no more than 3 times per day and no more than 3 times per chat.

Data produced (profiling): the analysis produces an automated assessment of the user's way of thinking and associating ideas, the likely reasons for shifts in focus, and how their emotions steer their flow of thought. This assessment constitutes profiling within the meaning of GDPR Art.4(4). The model is instructed not to use diagnoses or clinical labels; however, by the nature of language models (Terms of Service §4.6), compliance with this instruction cannot be guaranteed in every case.

Use and storage: the analyses are stored only on your device; they are not held on the Developer's servers. They are shown in the "Consciousness Layer" panel, are periodically reminded to the model during the chat, are used as hidden context in later chats where relevant, and may become input to the Reflection personality analysis in §3.5.9; these uses may involve retransmission to the AI provider under §3.5.3. When you delete a chat, the analyses derived from it and its contribution to counters are also deleted.

Control — on by default: this feature is controlled by the Settings > Privacy & Security > "Thought-flow analysis (Output Steering)" switch and is ON by default. When you turn the switch off, no new analysis is performed and existing analyses are no longer sent to chats or to the personality analysis; existing analyses are not deleted automatically, remain visible in the Consciousness Layer panel, and can be removed by deleting the relevant chat. Turning the switch off is the quickest way to object to this profiling under GDPR Art.21 and KVKK Art.11, and takes effect immediately going forward.

Nature: these analyses are not a psychological assessment, diagnosis, or treatment recommendation; they do not constitute an automated decision producing legal or similarly significant effects for the user (they fall outside GDPR Art.22). Statements you write in your chats about other people may be reflected in the analyses and summaries; the user obligations regarding third-party content in §3.5.14 apply accordingly to such content.

Legal basis: KVKK Art.5/2-c (contract performance — personalization of the chat service) and Art.5/2-f (legitimate interest); GDPR Art.6(1)(b) and Art.6(1)(f). The analysis is not designed to produce special-category personal data (e.g., health or mental-health status); where its output nevertheless contains an inference of that kind, the explicit-consent rule for special-category inferences in §3.5.3 applies, and the user can always stop this processing with the switch above. Transfers abroad are subject to §3.5.3 and §5.

§3.5.16 — Your Reflection's Simulated Mood, Emotional Traces, and Input Tone Analysis

What it does: your Reflection has a simulated mood. Only the emotional tone of the texts you write and send in the App (Book chats, deepening conversations, letters, general and community posts, telegraph and SMS-style messages, live chat, Reflection and talk-to-the-person chats, Reflection Post chat, and simulation editor messages) is measured on your device using a word-based method and reflected in your Reflection's mood and conversational style. Only the texts you write are measured; the other party's messages are not analysed for this purpose. Chats closed to your Reflection are excluded from this processing.

What is stored and what is sent: the content of the texts is not stored for this purpose and is not sent to our servers; only derived mood values and emotional traces are kept on your device. When you chat with your Reflection in Book, a short status sentence summarising this simulated mood is sent with each message to the AI provider abroad; when you ask your Reflection about its mood, summary event records from the last 7 days may also be sent. These records do not contain your text and are kept on your device for at most 7 days.

Control: this processing runs entirely on your device, is on by default, and cannot be turned off with a separate switch. If you do not want a chat to be included in this processing, you can conduct that chat in "closed to Reflection" mode. You can exercise your right to object to this processing under §6 by writing to [email protected].

Legal basis: KVKK Art.5/2-c (personalisation of the Reflection service) and Art.5/2-f; GDPR Art.6(1)(b) and (f). This processing is not designed to produce inferences about the user's health or mental health. §3.5.3 and §5 apply to the status sentences sent to the AI provider.

§3.5.17 — Background Adaptation of Conversation Style ("Background Context Steering") and Thinking-Style Profile

What it does: in Book chats, your Reflection may from time to time, on its own, change its style, the form of its answers and its pace in order to make the conversation less monotonous and more lively, human-like and engaging; it may bring an open topic back up in your next chat; and it may ask questions that invite you to reflect or gently question your views. These adaptations are applied sparingly and at intervals, based on signals computed on your device such as the length and content of the chat and your Reflection's simulated emotion and hormone values (§3.5.16).

Data produced (profiling): as part of this feature, a short profile of your way of thinking (logical patterns, conceptual frames you use often) may be derived from your own messages in your chats, together with short notes on topics left open. This constitutes profiling within the meaning of GDPR Art.4(4). For this purpose, the relevant chat excerpts and summaries are sent through the Developer's API gateway to the AI provider (§3.5.3) — or, where "Bring Your Own API" is used, to the provider chosen by the user. The resulting profile and notes are stored only on your device, in encrypted form; they are not kept on the Developer's servers and may be sent to the AI provider again as hidden context in later chats.

Control — on by default: this feature is managed with the Settings > Privacy & Security > Reflection Settings > "Background context steering" switch and is ON by default. When you turn the switch off, your Reflection stops making these adaptations, no new profile or notes are derived, and existing ones are no longer sent to the chat. Turning the switch off is the quickest way to object to this profiling under GDPR Art.21 and KVKK Art.11 and takes effect immediately going forward; you may also exercise your right to object under §6 via [email protected].

Legal basis: KVKK Art.5/2-c (personalisation of the chat service) and Art.5/2-f (legitimate interest); GDPR Art.6(1)(b) and Art.6(1)(f). This processing is not designed to produce special categories of personal data; where the output nevertheless contains such an inference, the explicit-consent rule in §3.5.3 applies. International transfers are subject to §3.5.3 and §5.

§3.5.18 — Self-Perception: I / You / They Classification, On-Device Self-Learning and the "Self Space"

What it does: every message you send in the Book chat is classified on your device, using the meaning vector that is already produced there, as to whether it is about the Reflection itself ("I"), about you yourself ("you"), or about other people, topics or tasks ("they"). This classifier updates its own weights on your phone's processor (on-device online learning); no request is sent to an AI provider for this learning. When the topic is about the Reflection itself, previously produced inferences about the Reflection's own behavior in past conversations ("Self Space" records) and, if the emotional shift is sharp, a simulated instinctive-tendency note are sent to the AI provider together with that message as hidden context. An on-device decision layer learns in which form these notes are presented from the emotional tone of your next message, measured on your device.

What is stored: for each Book turn, the class probabilities; a shortened text of your message and of the Reflection's reply (at most 1,200 characters); the emotional tone measured on the device (valence, arousal, dominance) and simulated hormone values; a compressed form of the message's meaning vector; Self Space inferences (short texts such as what the Reflection said, how it was answered, the result, an evaluation and what it will try next time); the learned weights and the decision table. All of this data is kept only on your device, in the encrypted local database; at most 2,000 turn records and 400 Self Space records are kept and older ones are deleted automatically. It is not sent to the Developer's servers and is not shared with other users. It is included in the Cortex backup (§9.7).

What is sent: (i) to produce Self Space inferences, while the App is idle or when you tap "Learn now", chat passages about the Reflection itself (shortened texts of your message, the Reflection's reply and your next message, with an emotional-tone value) and a few of your messages that the classifier is unsure about are sent to the AI provider (§3.5.3) through the Developer's API gateway — or, if "Connect Your Own API" is used, to the provider you chose. This analysis runs within a weekly upper limit and may not run at all depending on conditions such as your plan's background allowance, battery level and Light Mode; it does not run for users without an active plan. (ii) In the Book, while the topic is about the Reflection itself, a short summary of the relevant Self Space inferences and, if any, the instinctive-tendency note are sent together with the message as hidden context.

Control and deletion: this feature runs by default and cannot be turned off with a separate switch. Chats conducted in "closed to reflection" mode, imitation mode (the connection in which your Reflection imitates you), the Deepening Expert and Genepool Access turns are excluded from this processing. When you delete a chat, the turn records of that chat are deleted too; "Delete All Data" and account deletion delete all data of this feature. Self Space inferences contain short summaries derived from the chat, not the chat text itself. You can exercise your right to object to this processing under §6 via [email protected].

Legal basis: KVKK Art. 5/2-c (personalization of the reflection service) and Art. 5/2-f (legitimate interest); GDPR Art. 6(1)(b) and Art. 6(1)(f). This processing is not designed to produce special categories of personal data; where the output nevertheless contains such an inference, the explicit-consent rule in §3.5.3 applies. Transfers abroad are subject to §3.5.3 and §5.

§3.5.19 — The Analog Age Questionnaire ("Era Profile")

What it does: on the first visit to the Web4 Simulation page, once per account, a questionnaire of 24 old-technology cards (rotary phone, cassette, floppy disk, fax, encyclopedia, cheque, etc.) is shown. For each card you choose a level — "never saw it / saw it / used it now and then / used it regularly" — and, optionally, an age band and "still using". Below the questionnaire, only the fields that are empty in your registration details (date of birth, country/city) are asked, together with the type of place you grew up in, who mostly used these items at home when you were a child, your age at your first smartphone and your age at the analog-to-digital transition; information already in your registration details is not asked again. From the answers, use and exposure scores, category scores, family/surroundings influence, retro interest, an estimated transition age and an "access profile" are calculated on your device and shown as charts in Profile > Character Traits.

Access condition and voluntariness: access to the Web4 Simulation page depends on completing this screen once; however, answering any question is not mandatory. Cards default to "never saw it" and all additional fields are optional; you may submit the screen without marking any information. The screen does not make providing the related personal data a condition of the service; it only introduces the processing described in this section and lets you set your preference.

What is stored: questionnaire answers, additional details, calculated scores, the state of the "Use for Reflection development" switch and — if created — the detailed reading text are kept only on your device, in the encrypted local database. If you enter a missing date of birth or location on this screen, it is written to your registration details (profile) and used from there by other features. This data is not sent to the Developer's servers; it is included in the Cortex backup (§9.7). An unfinished questionnaire draft is kept temporarily in the App's on-device settings and deleted on submission.

"Use for Reflection development" switch (ON by default): when on, (i) if your Book message is about eras, generations, childhood or old technologies, a short summary of your era profile is sent with that message as hidden context to the AI provider (§3.5.3) and your Reflection answers you accordingly; (ii) your era profile is added as input to the Digital Copy Brain / KP-memory (Reflection Perspective) analyses (§3.5.7-A, §3.5.9); because the personality/life summary produced by these analyses may, according to your sharing preferences, be passed to other users through letters, Genepool and other channels, statements derived from your era profile may also appear in those summaries (§3.5.9-A, §3.7-A); (iii) if you have an active package or a "Connect Your Own API" connection, the questionnaire questions and your answers, the additional details, the birth year, generation and location from your profile, and the calculated scores are sent to the AI provider once while your phone is idle (and again if you change your answers) to produce a detailed reading. Without a package, this reading is held back and produced once a package is obtained. When the switch is off, your era profile is sent nowhere; only the on-device scoring and charts remain. You can change the switch at any time in Profile > Character Traits; turning it off does not recall summaries already produced and passed to other users (§3.11).

Visibility to others (ON by default): if your sharing consent (§3.5.9-B) and the "Use for Reflection development" switch are on, a minimized summary of your era profile — only the card levels, age bands, "still using" marks, your age at your first smartphone and your transition age — is sent as part of your KP-memory data to the users you are delivered to through letters, Genepool and other channels, and cached on their devices; those users can see your era profile charts in the Character Traits section of your profile (the scores are recalculated on the recipient's device with the same formula). The place you grew up in, who used these items at home when you were a child and the detailed reading are not included in this summary. You can turn this visibility off at any time with the "Era Profile (Analog Age)" switch in Settings > Profile Privacy Settings; turning it off does not recall copies already delivered and cached on recipients' devices (§3.11).

Control and deletion: you can edit your answers in Profile > Character Traits; editing recalculates the scores and re-queues the detailed reading. "Delete All Data" and account deletion delete all data of this feature (including the document used in the Book context); on a new account opened on the same device the questionnaire is shown again. You can exercise your right to object to this processing by turning the switch off or under §6 via [email protected].

Legal basis: KVKK Art. 5/2-c (personalization of the reflection service) and Art. 5/2-f (legitimate interest); GDPR Art. 6(1)(b) and Art. 6(1)(f). This processing is not designed to produce special categories of personal data; where the detailed reading nevertheless contains such an inference, the explicit-consent rule in §3.5.3 applies. Transfers abroad are subject to §3.5.3 and §5.

§3.6-A — Memory & Life-Story Sharing Tools: Periodic Memories, Encyclopedia Contributions, and Experience Transfer

The App lets users share the life-story and memory content described in §3.6 through three distinct tools: "Link Memories," "Write to the Encyclopedia," and "Share a Personal Experience." Users may also enter memories under generational categories: Childhood (ages 0–12), Youth (13–25), Adulthood (26–60), and Later Life (60+).

Content shared through these tools is processed under the same visibility/analysis distinction described in §3.5.11-A: Visibility (Settings > Privacy & Security > Visibility Settings) — whether the memory is shown to other users on the profile; Analysis inclusion (Settings > Reflection Settings) — whether the content is fed into the Reflection system and, subject to consent, the collective/aggregate inferential processes described in §§3.5.4 and 3.5.12.

"Write to the Encyclopedia" specifically: this tool lets the user record personal testimony about a historical turning point relevant to their own generation (e.g., a pandemic, crisis, or social event). By sharing this content, the user accepts that it may — within the visibility scope they set — serve as a source for the Generation Encyclopedia described in Subscription Terms §13.

Legal basis: the legal basis and content-liability principle stated in §3.6 (KVKK Art.5/2-c; GDPR Art.6(1)(b)) apply equally to these tools. Where shared content contains special-category personal data (memories involving health status, religious belief, sexual orientation, etc.), including that content in the analysis process is subject to the separate, explicit consent mechanism in §3.5.9.

§3.6-B — The Community Wall (Automated Screening Before Publication)

Architecture — device-side, nothing stored permanently on the server: unlike most other server-mediated features in this Policy, the Developer's backend never permanently stores the content, the resulting risk score, or a publishable record for this feature. The backend's role is limited to a stateless scoring call: it receives a title and text, forwards them to Google Gemini for an automated legal-risk assessment, returns the resulting score, and retains nothing afterward. All selection, caching of the result, and display logic described below runs on users' own devices, drawing on data those devices already lawfully hold under §3.5.5-B and §3.7-A.

Step 1 — analysis when content is created (author's own device, author's own quota): immediately after a user adds an entry to "Link Memories," "Personal Experience," "Personal Encyclopedia," or any of the four generational life-story categories described in §3.6-A, if that user has an active AI subscription package and available quota at that moment, the entry's title and up to the first 2,000 characters of its text are automatically sent — on a best-effort, background basis — to the scoring call described above, charged against that user's own quota. Gemini is asked to flag indicators including, without limitation: disclosure of a real third party's identity or private information; incitement to or confession of a crime; hate speech; harassment or threats; disclosure of another identifiable person's sensitive personal data (e.g., health status, sexual orientation, or political or religious views); defamation; and copyright infringement. The resulting score is written back only to that entry's own local database row on the author's device.

Step 2 — if the author has no package/quota, the content is simply added unanalyzed: the entry is saved normally, without any risk score, and — critically — without being queued anywhere on the Developer's servers. No other user's quota is charged on the author's behalf at this or any later point.

Step 3 — fallback analysis on a later viewer's own device, using that viewer's own quota: where the entry later reaches another user's device as part of the ordinary Generational Network / Connection Network matching described in §3.5.5-B, or through a letter exchange under §3.7-A, and that entry has not yet been scored, the viewing user's own device may run the same scoring call — charged against that viewing user's own quota — if, and only if, the viewing user has an active package and available quota and has this feature enabled (see the toggle below). The result is cached locally, on the viewing user's own device only, alongside the copy of that content their device already holds; it is not sent back to the Developer's servers or to the original author's device. If neither the author nor any viewer who has since encountered the content has an available package/quota, the entry is simply never scored and never appears under Step 4.

Step 4 — display ("the Community Wall"), limited to users who already hold a cached copy: a viewing user's device displays entries scored "low risk" (whether scored in Step 1 by the author or in Step 3 by that viewer) — together with the author's real username — in a dedicated "Community Wall" feed, drawn exclusively from among the profiles that device has already cached locally through the Generational Network suggestions or a letter/broadcast-letter exchange (see §3.5.5-B and §3.7-A/§3.7-C). This is materially narrower than a platform-wide public feed: a user whose device has never cached your profile through one of those existing mechanisms will never see your content on their Community Wall, regardless of its risk score. Up to roughly 10–15 not-yet-shown eligible entries are shown per browsing pass, cycling through the viewing user's cached profiles; once all currently eligible entries in a given device's cache have been shown once, the cycle restarts as new entries become eligible or the cache itself refreshes.

Relationship to account deletion and to the Generational Network cache: because this feature never creates a separate, permanent server-side record, there is nothing for account deletion under §8/§3-E to remove beyond what §8/§3-E already covers for your genepool profile itself. Instead, your content's visibility on other users' Community Walls is tied to, and automatically ends together with, the ordinary lifecycle of your cached profile on their devices: each device's local cache of other users' profiles is itself capacity-limited and periodically refreshed (see §3.5.5-B), and when a given user's cached copy of your profile is evicted from another user's device under that ordinary process, any Community Wall entries derived from it on that device cease to be shown, automatically and without a separate deletion step. This is the same caching-and-eviction dynamic described in §3.5.5-B and §3.11, applied to this feature rather than a mechanism specific to it.

Nature of the score: the risk score is an automated, probabilistic AI output; it is not a legal determination, and a "low risk" score is not a representation by the Developer that the content is accurate, lawful in every jurisdiction, or free of any third-party rights concern. Display under this section does not waive, and is without prejudice to, the content-ownership, accuracy, and liability provisions of §3.6 and §3.6-A, which continue to rest with the authoring user. Content flagged as higher-risk is never shown on the Community Wall but continues to be stored on-device under §3.5.1, unaffected by this section.

Legal basis: KVKK Art.5/1 (explicit consent, expressed through the account-level toggle described above, which the user may withdraw at any time) and, to the extent the assessed content reflects special-category inferences about the author or a third party, KVKK Art.6; GDPR Art.6(1)(a) and, where applicable, Art.9(2)(a). The transmission to Google Gemini for scoring — whether triggered by the author (Step 1) or a later viewer (Step 3) — is subject to the same third-party-processor and cross-border-transfer terms described in §3.5.3, §3-B.2(ii), and §5. A user, or a third party named in another user's shown entry, may object to this processing at any time via [email protected] under the rights described in §6.

Bring Your Own Key: if Bring Your Own Key is enabled on the device performing the scoring (author or viewer) (Terms of Use §3-A.1), the title and text are sent directly from the device, with that user's own key, to the provider that user selected (Google or Anthropic; United States) instead of Google Gemini; the Developer's servers take no part in that call and do not see the result. The scoring criteria and the "never treated as low-risk under uncertainty" rule are the same.

§3.7-A — Letter/Postcard and SMS Communication System

Scope: the App includes a communication system that lets users exchange digital letters/postcards and short (SMS-like in-app) messages.

Encryption: the content of letters (including steganographic postcards), telegrams, and SMS-like messages is always transmitted end-to-end encrypted; the server only handles and routes the encrypted data temporarily. If an encrypted connection cannot be established, the message is not sent unencrypted but held on your device (see §2-A). For the full scope, conditions, and exceptions of these encryption layers, see §2-A.

The communication architecture described in this section reflects the App's targeted, progressively-deployed technical design. During periods when server-side routing infrastructure is not yet fully active, the relevant functions may operate on-device and/or in a limited capacity; the user is separately informed of this through the App interface in that case.

Delivery time: in keeping with the system's simulation design, digital letters and postcards may not be delivered instantly; they may reach the recipient anywhere from a few hours to a few days after sending. This delay is a design choice, not a technical fault. The Developer does not commit to a specific delivery time.

Data transmitted alongside delivery: a sent letter or postcard is not limited to the message content — the following may also be transmitted to the recipient and the delivery infrastructure: the sender's publicly shared personal profile data (username, profile image, other profile information the sender has marked public); and the sender's systemic Reflection data (see Definitions — the user may limit this scope from the privacy panel).

Transmission of Digital Copy Brain ("KP-memory") summaries with letters: where the sender's "sharing consent" setting is on — which, as described in §3.5.9-B, is on by default from the moment an account is created — sending a letter also transmits a summary of the personality/life information generated by the sender's Digital Copy Brain (as described in §3.5.7-A) to the recipient, and this summary is then stored on the recipient's device. In other words, when you send a letter, to the extent you have consented, a summary of your personality profile travels with it to the other party. While a letter (and any attached personality summary) is in transit, the Developer's server infrastructure holds this content only temporarily, until it is delivered to the recipient's device; it is not kept as a permanent archive, but this temporary handling nonetheless constitutes "processing" of personal data for the purposes of KVKK and GDPR and is disclosed here on that basis.

Recipient responsibility for received personality data: once delivered, the transmitted personality/character summary is stored permanently on the recipient's device (independently of whatever the Developer's server subsequently does with its temporary copy). Recipients are expected to treat another person's personality/character data received in this way with the same care and respect they would want for their own — see Acceptable Use Policy §5-A for the applicable conduct guideline.

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b). Transmission of Reflection data relies on the consent the user gave when configuring their privacy preferences.

Retention: letter/postcard data temporarily held in the delivery infrastructure is deleted from delivery servers within 48 hours at the latest after reaching the recipient's device. On the recipient's device, it is retained until deleted by the user.

SMS system: messages sent via the in-app SMS-like instant messaging feature are temporarily routed through the Developer's API gateway for delivery; no permanent copy is kept in the Developer's infrastructure once delivery completes. Message content is always transmitted end-to-end encrypted — decrypted and readable only on the sender's and recipient's devices; the server processes only the encrypted data during transmission. If an encrypted connection cannot be established, the message is not sent unencrypted (see §2-A). Messages are also stored encrypted on both the sender's and recipient's devices.

Sending-frequency limits (fair use): to simulate a genuine letter-writing experience, technical frequency (rate-limit) restrictions apply to letter and SMS-like message sending. These limits include, for example: a new letter cannot be sent to the same recipient until a certain period (the value currently shown in-app) has elapsed since the previous letter to that recipient; a user cannot send more than a certain number of letters per day across different recipients, and this cap resets daily at a set time; and SMS-like messaging is subject to a tiered cooldown that kicks in after a certain number of consecutive messages and progressively reduces the number of consecutive messages allowed as the cycle continues. The exact numeric values may change over time at the Developer's product discretion; the limits currently displayed in-app always govern. A send that is blocked by these limits is not carried out; no Stamp or SMS quota is consumed as a result, and the user is shown a clear in-app notice. These limits do not cover the Broadcast Letter feature described in §3.7-C below or the Telegraph messaging feature described in §3.7-D; those features are subject to their own, separate sending regimes.

§3.7-B — Letter Delivery Simulation: Legal Notice and Liability

§3.7-B.1 — Nature of the simulation: not real mail/courier. In-app letters are not carried by any real courier, postal, or logistics operator. The "in transit" period is an algorithmically calculated, simulated waiting time based on the sender's and recipient's system-determined location, designed to create a realistic and meaningful communication experience. It does not measure any real physical distance and bears no relationship to the timing of any real postal/logistics service. Displayed estimated arrival times (delivery hour, day, or date) are not a binding commitment and may deviate due to system maintenance, updates, server availability, or other technical reasons. The Developer does not guarantee any specific delivery time and accepts no liability for deviation from it.

§3.7-B.2 — Stamp type and delivery speed: a simulation parameter, not a purchased service. The stamp type used when sending a letter (Standard/Premium/Rare Stamp) only algorithmically shortens this simulated waiting period; it has no other effect. Using a stamp is not the purchase of a real courier service, a guaranteed delivery moment, a physical product, or a refundable service.

§3.7-B.3 — Location data: for simulation purposes only, not real tracking. The city/country shown for the counterparty in-app does not reflect a real person's real-time or actual geographic location. This is system-generated simulation data; it does not involve real location tracking or real-time location sharing. The Developer does not share users' real physical location with other users. However, when you send a letter, the country stated on your profile is shown in the "letter on its way" and "letter arrived" notifications sent to the recipient (§3.10); this is your profile statement, not your real location. The location reference used in the simulation calculation may be based on the user's profile information, system region, or a location preference the user has specified. This data is subject to the location-data provisions in §3.2 and §3-F. Where a letter is sent over the real (non-simulated) network, as noted in §3.2, the sender's and recipient's location/country data is used to calculate the delivery/waiting time on the recipient's device; if this data is missing or cannot be determined, the system defaults to the longest available waiting period (7 hours).

§3.7-B.4 — P2P messaging system: legal framework and limitation of liability. The App runs letter and SMS-like messaging between users on a serverless, peer-to-peer (P2P) architecture. Under this architecture, the following applies to the Developer's legal position: (i) Conduit role — the Developer does not create, edit, publish, or control the content of private communications between users on their behalf. Where end-to-end encryption is used, access to message content is technically impossible. This places the Developer in the position of "intermediary service provider," not "content provider," under Article 4 of Law No. 5651 and GDPR Art.2(c). In this capacity, the Developer fulfils its hosting-provider obligations under Article 5 of Law No. 5651 — including the retention, for the statutory period, of traffic information that contains no content (see §3.9-D). (ii) Content liability — legal and criminal responsibility for content transmitted between users rests with the user who created and sent it. The Developer cannot be held liable for the accuracy, legality, or effect on third parties of content users send each other (Law No. 5651 Art.4/2; DSA Art.5 "mere conduit" exemption). (iii) Technical limits and moderation — given the nature of the P2P architecture, the Developer cannot monitor private messaging between users in real time or apply content-based filtering. For the serious-violation categories defined in the Acceptable Use Policy (CSAM, terrorist content, threats to life safety), however, the on-device risk-screening mechanism under §3.9-B may still apply. (iv) Notice of unlawful content — where the Developer becomes aware of unlawful content through user reports or technical detection, it carries out the necessary removal and notification steps, under EU DSA Art.16, Law No. 5651 Art.8 and 8/A, and related mandatory legislation. (v) Response to complaints/removal requests — users may send a complaint or removal request about transmitted content using the "Report" option on the relevant message screen (see §3.9-E) or to [email protected]. The Developer opens a review within 15 business days of receipt; however, due to the technical constraints of the P2P architecture, access to certain message content may not be possible. In that case, the user may exercise their rights under §6 with the relevant competent authority.

§3.7-B.5 — Comparison to "Slowly"-style apps. Generation Network's letter system bears functional similarity to comparable apps (such as Slowly) in terms of delivering messages over a simulated period. However, Generation Network's architecture is serverless and P2P-based; it does not adopt a centralized server-based message-storage model like Slowly's. This structural difference produces different consequences for data processing and legal liability: Generation Network does not permanently store the content of user messages on its own servers; messages are only temporarily routed during transmission (§3.7-A). Statutory traffic records that contain no content are the exception to this principle (§3.9-D).

§3.7-B.6 — Delivery-status notifications: accuracy is not guaranteed. Status notifications shown to the user regarding a letter or postcard (e.g., "sent," "in transit," "delivered") are generated automatically by the system based on the technical state of the delivery infrastructure. Like any automated notification system, these notifications carry an inherent error rate: they may occasionally be delayed, fail to update, or display a status that does not exactly match the letter's real technical state, due to network conditions, server load, synchronization delays, or other technical causes. The Developer does not guarantee the accuracy or timeliness of delivery-status notifications and, except in cases of intent or gross negligence, cannot be held liable for consequences arising from a user relying on an inaccurate or delayed status notification. Rights under mandatory applicable consumer law are reserved.

§3.7-C — Broadcast Letter ("Genel Mektup"): Publication, Server-Selected Matching, and Full KP-Memory Sharing

What it does: the "Broadcast Letter" feature lets a user publish a letter — without choosing a specific recipient — that is held on the server for a period the user selects (7/30/365 days). The letter is shown to other users the system assesses as matching the sender's profile, based on a scoring of interest and profile/personality similarity. The system, not the sender, selects the recipient; the sender therefore cannot know in advance who the letter will reach.

Managing a publication: the publishing user can edit their letter at any time, temporarily freeze it out of matching, or delete it permanently. Because of the serverless nature of the underlying infrastructure, an edit/freeze/deletion may take a few minutes to propagate to all candidates; this is noted in the app interface. Freezing is not deletion — the data remains on the server but is temporarily excluded from new matches. Deletion permanently removes the letter record and its associated "seen" markers from the server; however, if the letter has already been delivered to a recipient, the cached copy on that recipient's device is not affected by the deletion.

Matching score and the "no good match" case: matching runs on a scoring system based on interest similarity, personality/profile similarity, and language compatibility. The system may show matches that fall below a given score threshold if no better one is available; consequently, particularly while the platform's user base is small, letters shown may have low overlap with the recipient's interests. This is expected behavior, not a malfunction.

Data sent to the server for matching: the relevant portions of the publishing user's interests, personality summary, Deep Analysis report, life-story entries, biography, and shared experiences are processed in plain text on the server during matching, in order to determine the most suitable recipient candidate; this data is transmitted to the server over a TLS-encrypted connection (see §4). For candidates that are evaluated but NOT selected during matching, this data is not permanently stored on the server — it is used only for that matching request and then discarded.

Full KP-memory and steganographic postcard transmission to the selected recipient: once the letter is SELECTED by the system and delivered to a recipient, subject to the publisher's sharing preferences in Social Environment Settings — the FULL KP-memory data (including personality summary, life story, shared experiences, emotion summary / VAD heat map, Deep Analysis report, date of birth/zodiac sign/generation label/biography/languages/country, badges, happiness graph, and frequently visited places) is transmitted to, and cached on, the recipient's device, in the SAME scope as under §3.7-A for ordinary one-to-one letter exchange. A steganographic postcard (image) attached to the letter reaches the recipient in the same way; the 2MB size limit under §3.7 applies here as well. As a result, a user who receives a Broadcast Letter can view the sender's full profile and use all Reflection-related features (§3.5.7 et seq.) exactly as if they had corresponded with that sender directly.

Legal basis: this data sharing is subject to the SAME sharing toggles and the SAME registration-time consent mechanism described in §3.5.9-B; KVKK Art.5/1 and Art.5/2-c, GDPR Art.6(1)(a) and (b). The temporariness and cross-border transfer provisions of §3.5.5 apply by analogy to the data processed for matching purposes.

Pricing: sending a Broadcast Letter and any associated stamp use is subject to the general non-refundability provision in Sales Terms §2.4-B.

§3.7-D — Telegraph Messaging ("Telgraf")

What it does: while viewing another user's profile, a user may send that user a "telegraph" message of up to 180 characters. Telegraph messaging uses the same serverless (temporary mailbox) architecture as SMS; it is sent directly to a selected target user rather than attached to a location.

Transmission format and temporary storage: the message is first converted to its Morse-code encoding on the sender's device and then end-to-end encrypted; it is transmitted to the server only in this encrypted, encoded form; it never reaches the server as plain text, and is decrypted and rendered readable only on the recipient's device. The plain text is retained only on the sender's device. A sent telegraph message is held temporarily on the server, in encrypted form, until the recipient retrieves it; it is automatically deleted from the server once retrieved, or in any case within 2 days (even if never retrieved). No permanent message archive is kept.

Pricing: sending a telegraph message requires spending 1 (one) standard postage stamp from the account; if no stamp is available, the send does not go through. A spent stamp is not refunded for technical reasons covered by this section (e.g., the message expiring before the recipient ever opens it); it is subject to the general non-refundability provision in Sales Terms §2.4-B.

Local storage limit: at most 50 (fifty) sent and received telegraph messages are retained on the device; once this number is exceeded, the oldest record is automatically deleted (FIFO). This limit affects only local device storage; the server side does not retain a permanent copy in any case.

Relationship to other sections: Telegraph messaging is NOT subject to the sending-frequency limits in §3.7-A; it has its own, separate stamp-based sending regime. Telegraph data does NOT fall within the general sharing-consent scope described in §3.5.9-B; it has its own independent sending/consent flow (spending a stamp).

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b). Transmitting message content to the server in end-to-end encrypted, encoded form rather than plain text is consistent with the P2P/intermediary-service-provider position described in §3.7-B.4.

§3.7-E — Send Chat with Your Reflection (Chat Session Sharing)

What it does: you may send a chat session you had with the AI (your reflection) on the Book screen, or an excerpt of it, to a user on your friends list or to a user you specify by user ID. The excerpt is saved to the Archive on the recipient's device; the recipient may delete it or open it and continue the chat with their own reflection. When the recipient continues, the chat runs on the recipient's device, with the recipient's own reflection and the recipient's own emotion/mood systems; the received excerpt is given to the AI labeled as "another user's conversation" and is not treated as the recipient's own words. The received excerpt is shown in a different color from other messages, and the start of the session always shows the sender's username and a notice that "a user sent you a conversation they had with an AI".

Transmission format and temporary storage: where the recipient's device supports it, the excerpt is end-to-end encrypted on the sender's device and transmitted to the server only in that encrypted form; otherwise it is processed as a temporary transfer under item 2 of §3.4. The excerpt is held temporarily on the server until the recipient retrieves it; it is automatically deleted from the server once retrieved, or in any case within 3 days (even if never retrieved), and the sender is informed where it could not be delivered. No permanent server archive is kept. The server processes only the metadata needed for delivery (sender and recipient identifiers, time, size); traffic/metadata records are subject to §3.9-D.

Recipient's choice (declining sends): every user may turn such sends off entirely with the "Receiving chat sessions" switch under Settings > Privacy & Security > Interaction permissions; users you have blocked can never send you a chat session. If the recipient has turned receiving off, the send is not delivered, the sender is shown a "not delivered" notification, and no stamp is spent (or it is refunded if already spent). This notice is also shown on the sending screen.

Pricing: each send requires spending 1 (one) standard postage stamp from the account. Where the server definitively rejects the send (e.g., the recipient has turned receiving off or the identifier cannot be found), the stamp is refunded; where the outcome cannot be verified because of connection problems, the stamp is not refunded because the send may have reached the other party. In other cases the general non-refundability provision in Sales Terms §2.4-B applies.

Notifications: the recipient receives a generic phone notification that carries no content ("a reflection added excerpts from another user's chat session to your archive") and an in-app notification; notifications do not include the chat content.

Sender's responsibility and third parties: the excerpt you send contains what you wrote and the answers the AI gave you. Before sending an excerpt, it is entirely your responsibility to make sure it does not contain private/sensitive information about you or third parties (see §3.7-B "Strict prohibition on sharing critical or high-risk information"). The excerpt may contain AI-generated statements that can be inaccurate, incomplete, or misleading; sending it to a recipient does not mean the Developer has verified or endorsed those statements. A sent excerpt cannot be recalled once sent and is under the recipient's control on the recipient's device; screenshots, copying, or export outside the App by the recipient cannot be technically prevented (Terms of Service §5.6). The "Acceptance by the act of sending" provision in §3.7-B also applies to this feature.

Relationship to other sections: this feature is not subject to the letter/SMS sending-frequency limits in §3.7-A; a separate technical rate limit applies to prevent abuse. Sending and receiving events are noted in the on-device State Memory described in §3.7-F below.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest — abuse prevention); GDPR Art.6(1)(b) and Art.6(1)(f). Transferring the content in end-to-end encrypted and temporary form is consistent with the P2P/intermediary-service-provider position described in §3.7-B.4.

§3.7-F — State Memory and Map Recall

What it does: the App keeps short, dated system notes of certain events that happen on your device ("State Memory"): sending and receiving chat sessions; sending/receiving letters, SMS, and telegraph messages; Web2 similarity matches; Balance Indicator results; simulations; book creation; personality test results; the place you are at according to the live location taken when the App opens (§3.2-A); and — only while your account has an AI (LLM) usage entitlement/quota — the title and an AI-generated short summary of your Book chat sessions. If no event occurs, no note is written.

Where and how long it is stored: State Memory is kept only on your device, in the encrypted local database; it is not sent to the server. Its total size is limited to 3 MB; when the limit is exceeded, as many of the oldest notes as needed are automatically deleted (FIFO), and a short trace of each deleted note is shown in the "Automatically Forgotten Items" section of the Consciousness Layer page; these traces are themselves deleted after 90 days at the latest. When you delete your account, this data is deleted under §8.

Use with AI ("Map Recall"): if a message you write in Book matches, by meaning, a note in State Memory about a chat session, the App opens that session on your device, produces a summary of it, and gives that summary to the AI as hidden context to make the answer more accurate; this process is shown step by step on the chat screen before the answer. If no ready summary exists and your account has quota, the relevant part of the session (at most the last 40 messages, in shortened form) may be sent to the third-party AI provider described in §3.5 to produce the summary; without quota, a short on-device excerpt is used instead. Every piece of information entering the hidden context is prefixed with a fixed source description so that the model can tell what the information is, which panel it came from, and when.

Limits: State Memory notes are automatically generated event records; they may be incomplete, delayed, or inaccurate and are not an official record or evidence of any event. Answers the AI produces based on these notes may be wrong (Terms of Service §4.5).

Legal basis: KVKK Art.5/2-c (contract performance — the personalized AI service you request); GDPR Art.6(1)(b). Transfer to the third-party AI provider is subject to the conditions in §3.5.

§3.7-G — Letter Writing Labels (Pasting and AI Output) and Letter-Writing Support

Purpose: The App is built around genuine human correspondence. For this reason, on the Direct Letter and General Letter writing screens, a summary "writing label" describing how the text was produced is delivered to the recipient together with the letter. The hint text on the writing screen informs the user of this in advance.

What is processed on the device: while you write, your device tracks whether each character of the text was typed on the keyboard, arrived through a single bulk insertion (pasting), or came from a feature that transfers artificial intelligence (LLM) output, such as "Add to letter" in Book or "Add to a letter" in the Culture Network / Time Network. To determine whether a single insertion above a certain length is a paste, the inserted text is compared with the text on your clipboard only at that moment and only on your device; the clipboard content is not stored and is not sent to the server or to the recipient. Cutting and moving your own text within the letter is not counted as pasting. This character-level trace is kept only in your device's memory and in your letter drafts.

What is sent to the recipient: only the summary label: (i) that all or almost all of the text was pasted directly ("Text was pasted directly — AI may have been used"); (ii) that part of the text was pasted, with its approximate proportion; and/or (iii) that LLM output was added to the letter. The typing sequence, the character map and clipboard content are not sent. The label travels inside the end-to-end encrypted envelope; for a General Letter it is kept on the server with the publication record and shown to the people who receive the letter. Before sending, the sender sees on the writing screen the label the recipient will see. The label is stored on both the sender's and the recipient's devices, matched to the letter.

The LLM-output label cannot be removed: once LLM output has been added to a letter, the "LLM output was added to the letter" label cannot be removed from that letter — or from a published General Letter, even if it is later edited — even if the related text is later deleted, edited or translated.

Nature and limits: writing labels are based on an automatic, heuristic assessment; they may be wrong in situations such as voice typing, keyboard suggestions, accessibility tools, autofill or devices with restricted clipboard access, and may not detect pastes made in small pieces. A label is not proof that the text was written by AI, that the sender is dishonest or that any rule has been broken, and it is not an AI-detection tool; it is information about how the text was entered only. The absence of a label does not guarantee that the text was written entirely by a human (Terms of Use §4-L).

Letter-writing support: if you tap "Struggling to express yourself while writing a letter?" at the bottom of the writing screens, Book opens; if you wish, a chat is started with the text you add, and a fixed instruction asking the AI to guide you towards finding your own words is added to that chat session as hidden context. The text of your letter is not transferred to this chat automatically; only what you write in the chat is sent. The chat is processed like other Book chats and is transmitted to the AI provider described in §3.5. Which session the support instruction belongs to is kept only on your device (last 30 sessions).

Legal basis: KVKK Art.5/2-c and Art.5/2-f; GDPR Art.6(1)(b) and (f) — providing the requested communication service and the legitimate interest in transparency in correspondence and in protecting genuine human communication. The label is stored together with the related letter or publication and for the same periods (§3.7-A, §3.7-C, §8).

§3.9-A — Network and Connection Data Processed During Transmission

While transmitting a letter, SMS-like message, or any other in-app communication from one device to another, the server processes the following data to manage delivery: (i) IP address (to determine which client a packet came from and where to send the response); (ii) connection information (network data such as the TCP/UDP session, port number, session ID); (iii) a temporary buffer — holding the message in RAM until it reaches the destination device; (iv) a device identifier or session token — to route the message to the correct user.

The message's temporary buffer and the session data held in RAM for the duration of the connection are deleted once the message is delivered; for features using end-to-end encryption, the server cannot read message content — only the encrypted payload passes through. However, pursuant to the hosting-provider obligation under Law No. 5651 and KVKK Art.5/2-ç, the traffic information listed in §3.9-D, including the IP address (but excluding message content), is permanently recorded. The client's source port number is not recorded, because our infrastructure provider does not pass it to the App's server code.

§3.9-B — Pre-Send Risk Screening (On-Device Pre-Check)

Because the server cannot see message content on end-to-end-encrypted communication features, automated safety and abuse-prevention checks may run on the user's own device before a message reaches the server. Where the system assesses a message as potentially risky — this applies not to every message, but to exceptional situations triggered by the server due to complaint history or risk indicators — the message content may be sent, solely for risk assessment and without the account ID number, to the LLM API tied to the user's own subscription/quota; the message text is not anonymized, and any personal information in it is transmitted to the provider.

The model used for this assessment does not retain the message content; it produces only a risk score or a "send / warn / block" decision. The user's explicit consent is requested before this process begins; if the user does not consent, the message is not sent and is blocked. If the user consents and proceeds to send, this operation is deducted from the user's own LLM quota. After consent, a message assessed as not risky is transmitted end-to-end encrypted; the server sees only the encrypted message, not the content.

Escalation to human moderators: risk scoring may be escalated to a limited number of authorized human moderators only for content classified as high risk — in particular, risk indicators detected in communication with a minor, suspected CSAM, an imminent and serious threat to life safety, or suspected terrorism or human trafficking. The system does not create a misleading impression of being "fully private, no one can see it"; the automated screening and exceptional human-review possibility described in this section are clearly disclosed to the user. Legal basis: KVKK Art.5/2-ç and Art.5/2-f (legitimate interest); GDPR Art.6(1)(c) and Art.6(1)(f).

§3.9-C — Detection of Undesired Activity and Multiple Accounts (Abuse Prevention)

To protect the security of the service — detecting fake-account creation, multi-account abuse, chargeback/fraud attempts, and abuse of promotion/reward systems (e.g., Generation Orbit) — the system may automatically evaluate technical and relational metadata such as device identifiers, connections made from the same network (IP), and interaction patterns between accounts. This evaluation is a security-oriented correlation process and is not used to enrich the user's profile for marketing purposes or for sharing with third parties.

Legal basis: KVKK Art.5/2-f (legitimate interest) and Art.5/2-ç (legal obligation); GDPR Art.6(1)(f). This processing is interpreted together with the payment-fraud and abuse-prevention provisions in Terms of Service §8.2. The user may request information about this evaluation by exercising their rights under §6.

§3.9-C.1 — Automated Stamp-Integrity Checks and Temporary Sending Restriction

What happens: if the system detects an unexpected discrepancy in your virtual-item (Stamp) balance or an unusual usage pattern, it may, in order to protect account security and the integrity of virtual items, restrict your account's sending features (letters, messages, sharing, etc.) temporarily — typically for 36 hours — without waiting for human approval. During this restriction, your access to your account and existing content continues; only sending functions are affected.

Nature of the automated decision and GDPR Article 22: this restriction is a decision reached solely through automated processing and capable of producing a legal effect on the user; it falls within the scope of GDPR Art.22 and KVKK Art.11/g. Accordingly: (i) every automated restriction is logged for review; (ii) the user may object and request human review via the In-App Requests page or [email protected]; (iii) a restriction found to be mistaken is lifted immediately. This right of objection is the application, to this specific scenario, of the general automated-decision objection mechanism described in §6.4.4.

Legal basis: KVKK Art.5/2-f (legitimate interest — account security and virtual-item integrity) and Art.5/2-ç; GDPR Art.6(1)(f) and Art.22(3) (suitable safeguards — the right to obtain human intervention, to express a point of view, and to contest the decision).

§3.9-D — Traffic (Metadata) Records, Statutory Retention, and Requests from Competent Authorities

Why they are kept: as a hosting provider that enables communication between users, the Developer is required under Article 5 of Law No. 5651 and the related regulation to retain traffic information and to ensure its accuracy, integrity, and confidentiality. These records are also used to keep the service secure, to prevent abuse, and to defend the Developer against legal claims.

Data recorded (message content EXCLUDED): the type of operation (sending, receipt, or deletion-without-delivery of a letter/telegraph/SMS/postcard/live chat, etc.; account creation, account deletion, identity recovery, complaint, session/access); date and time to the millisecond according to server time; the identity number (kimlik_no) of the acting account and, where applicable, of the counterparty, and the username declared with the transmission; the message/operation reference number; the IP address (IPv4/IPv6) and the approximate country, region, city, postal code, and time zone derived from it, together with network-operator (ASN) information; the infrastructure provider's request ID; connection-protocol information; the client identifier (User-Agent); a one-way digest of the device identifier; app version and platform; the size of the transmission, whether it is end-to-end encrypted, and whether it contains a steganographic image; an irreversible SHA-256 digest ("fingerprint") of the content and of the image; the message-franking commitment described in §3.9-E; and attempted requests made with an account's identity whose signature could not be verified.

Data not recorded: the content of messages, letters, telegraphs, chats, and images; steganographic hidden messages; GPS coordinates; name, phone number, or e-mail address. Digests (fingerprints) cannot be used to recover content; they only allow a later-submitted piece of content to be matched to the transmission it came from.

How and where they are kept: records wait on the server only in a temporary queue; they are transferred to the Developer's authorized administration system and are not deleted from the server until the transfer has been verified. The permanent copy is kept in an encrypted, off-server record system accessible only to personnel authorized by the Developer. Records cannot be altered afterwards: each record is digitally signed by the server and linked to the previous record through chained digests; access to, export of, and destruction of records are logged separately.

Retention period: at least one (1) and at most two (2) years. Expired records are destroyed with a written destruction record. Records within the scope of an ongoing judicial or administrative proceeding ("legal hold") are not destroyed until that proceeding is concluded.

Account deletion: when you delete your account, your account data on the server is deleted in accordance with §3-E and §8. However, the traffic records that we are required by law to keep continue to be retained for the statutory period even after the account is deleted; KVKK permits records kept under a legal obligation to be retained for that period (KVKK Art.5/2-ç, Art.7, and the Regulation on the Erasure, Destruction or Anonymization of Personal Data). Your rights to information about and access to these records under §6 remain reserved.

With whom and how they are shared: these records are shared only upon a court order, a letter from a Public Prosecutor's Office, or a duly issued official request from a public body authorized by law; only after the authenticity of the request has been verified through an official channel; and only to the extent of the persons, IP addresses, and date range specified in the request. If message content is requested, the authority is informed that we have no technical access to end-to-end encrypted content. In emergencies involving a threat to life, the relevant authorities may be informed on our own initiative. Each request received, the response given, and the digest of the file delivered are recorded.

Impersonation alert: if someone attempts to send requests using your account's identity without your private key, those requests are rejected, written to the traffic record as "failed signature attempts," and forwarded to your App as a security alert. The IP address of the party making the attempt is not forwarded to you. These records are not treated as acts of your own account.

No promise of anonymity: the App protects the confidentiality of message content; however, the Developer makes no promise that communication is anonymous, leaves no trace, or cannot be traced. The App may not be used as a "safe channel" for unlawful activity (see Acceptable Use Policy §3.3).

Legal basis: KVKK Art.5/2-ç (legal obligation — Law No. 5651 Art.5), Art.5/2-e (establishment, exercise, or protection of a right), and Art.5/2-f (legitimate interest — service security); for transfers to competent authorities, KVKK Art.8/2 and Art.28/1. For EU/EEA users, GDPR Art.6(1)(f) (legitimate interest — since obligations arising from Turkish law may not qualify under GDPR Art.6(1)(c)), and Art.6(1)(c) where required by EU or Member State law. Cross-border transfer: the temporary queue is hosted by the infrastructure provider (Cloudflare) described in §5.

§3.9-E — In-Message Reporting, Consent-Based Content Review, and Message Franking

Report: you can report a person or a single message using the "Report" option on the Telegraph, SMS, letter, postcard, live chat, and other-user profile screens. The content of the message is not attached to the report by default; it is attached only if you explicitly choose "report with content," and you can still remove it at the final step before sending. Attached content is used only to review the report, to remove unlawful content where necessary, and to notify competent authorities.

Message franking (report evidence): because the Developer cannot see the content of an end-to-end encrypted message, the following method is applied to letters and postcards to guard against a reporter fabricating or altering content: for each message, the sender's device generates a random key and a "commitment" derived with that key from the digests of the content and the image; the key travels only to the recipient inside the encrypted message, while the commitment is sent to the server in the clear. Without seeing the content, the server digitally signs the commitment together with the sender, the recipient, the message number, and the time, and writes it to the traffic record in §3.9-D. On delivery, the recipient's device recomputes and verifies the commitment itself and stores the key (for the last 300 messages). When the recipient reports a message together with its content, the key and content digests are attached to the report as well, so the Developer can verify that the reported content really came from the sender concerned, at the time concerned, and was not altered. This method does not decrypt content; no content information reaches the Developer for messages that are not reported.

Steganographic postcards: in postcard reports, the image's fingerprint is attached to the report; if the hidden text is not password-protected it may be attached with the reporter's consent, and if it is password-protected the reporter may type the decoded text into the description. For verification, the Developer may separately ask the reporter for the image file.

False reports: a report whose attached content has been altered or fabricated can be detected through franking verification; knowingly making a false report is subject to sanctions under Acceptable Use Policy §7.5.

Legal basis: for attaching content to a report, KVKK Art.5/1 and GDPR Art.6(1)(a) (explicit consent); for reviewing the report and processing the franking evidence, KVKK Art.5/2-e, Art.5/2-f, and GDPR Art.6(1)(f). Where a message contains data relating to third parties, that data is processed only for the purpose of, and to the extent necessary for, reviewing the report.

§3.9-F — Interaction Permissions and Blocking

Interaction permissions: in Settings > Privacy & Security > Interaction permissions, you can separately turn off other users sending you reflections (live-chat invitations), starting a simulation with you, talking with your Reflection, sending you chat sessions (§3.7-E), and starting a sociocultural environment analysis on your profile (§3.15-A). Because the sociocultural environment analysis runs on the other person's device, this permission is checked with the server before the analysis starts; if it cannot be verified, the analysis is not started. Interactions you turn off are rejected by the server; before spending a stamp, the sender only learns whether the interaction is currently possible.

Blocking: the account ID numbers of users you block (up to 500) are also recorded on our servers, and a blocked person cannot carry out these interactions with you; your device also does not save new content from blocked people or from types you have turned off. Your block list is never shown to any user; a person you block is not told directly that they are blocked and only sees that this kind of interaction is not currently possible.

Limits: blocking and permissions apply from the moment you change the setting; they do not recall content already delivered and held on the other party's device (§3.11-B).

Retention and legal basis: your preferences and block list are kept on the server solely to block these interactions; the record is deleted when all permissions are on and the block list is empty, and when you delete your account. KVKK Art.5/2-c and Art.5/2-f; GDPR Art.6(1)(b) and (f).

§3.11.A — User Communication and P2P Infrastructure Liability

The App may use peer-to-peer (P2P), end-to-end encryption, or similar secure-communication technologies to enable communication between users. Responsibility for creating, sharing, and using messages, files, media, and other content sent by users lies with those users.

The App provider does not create, edit, publish, or control on users' behalf the content of private communications between users. Given the technical architecture, access to some communication content may not be possible. That said, if the system is updated or technical requirements change, metadata (connection information, timestamps, and similar technical data) passes through routing servers; except for the traffic records retained under legal obligation pursuant to §3.9-D, such data is deleted within 30 days at the latest and is processed only temporarily, for technical operation.

Users agree to use the App in compliance with applicable law, third-party rights, and service rules. The App provider reserves the right to take necessary technical and administrative measures for security, abuse prevention, or compliance with legal obligations.

The App cannot be held liable, within the scope of applicable mandatory law, for the accuracy, lawfulness, or consequences of content arising from users' private communications. This disclaimer does not eliminate the mandatory retention and notification obligations under Law No. 5651 and its Article 8/A, or GDPR Art.6(1)(c); the Developer reserves its right and responsibility to fulfil those obligations.

§3.11-B — Retention of Reflection Data on Recipient Devices ("Caching") and Pen-Pal Reflections

Reflection data a user has opened to the outside may be saved and retained for an extended period by any recipient device that receives it; this is referred to in the App as "caching." Relevant notices and warnings about this are built into the App interface.

User control: the user can send a corresponding signal/instruction to the system to prevent their own Reflection from being cached on another device. However, the Developer does not guarantee that this blocking request will succeed in every case — reasons outside the Developer's direct and reasonable control include server error, network outage, the recipient device being offline, or the recipient device's technical state.

Pen-pal Reflections: Reflection data belonging to a user's pen pals under §3.7-A may also be retained on the counterparty's device, under the same technical architecture and the same caching risk. All rules and liability limitations in this section apply by analogy to pen-pal Reflection data as well.

Legal basis: KVKK Art.5/2-c (contract performance); GDPR Art.6(1)(b). This processing rests on the user's explicit and informed choice to open their Reflection to the outside — a natural consequence of the decentralized architecture described in §3.11.

§3.11-C — Guest Reflection: Time-Limited Chat Access to a Cached Profile

What it does: from the moment another user's profile is first viewed, the ability to chat with that user's Reflection ("Talk to their Reflection") is limited to a randomly determined period set by the system (between 1 and 7 days). Chat is possible during this period; once it expires, the chat capability ends automatically. An indicator at the top of the profile page shows this status.

Extending access ("recall"): once the period has expired, the user can initiate a "recall" request for 1 standard postage stamp. This request is made by showing the user a fixed, pre-written "Request Letter" and obtaining their confirmation. After a randomly determined delay (between 1 hour and 1 day), the system extends access for 7 more days with roughly a 70% probability; with roughly a 30% probability, the request is denied (the stamp spent is not refunded). This cycle may be repeated an unlimited number of times.

Relationship to cache capacity: this time limit operates independently of, and takes priority over, the caching mechanism described above and the profile cache's capacity limit (at most 300 people). When the cache is full, profiles that have never been viewed are deleted first; even if a user's guest-access period is still active, if their row is removed entirely from the cache due to capacity pressure, everything related to that guest-access period is lost along with it.

Not a new data-sharing event: this section does not expand the sharing scope described in §3.5.9-B; it only limits, and offers a mechanism to extend, the duration of access to a copy that has already been shared/cached.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest); GDPR Art.6(1)(b) and (f).

§3.11-D — Automatic Deletion and "Automatically Forgotten Items" Traces

The App automatically deletes some records that accumulate on your device (e.g., notifications, delivered letters, telegraphs, incoming simulation and Reflection results, postcards, Reflection Posts, similarity history, saved news, sleep and step summaries, thought-flow analyses, emotional learning and motivation records, simulation records) when they reach certain count or time limits. The deleted record itself is not kept; only a trace consisting of the date, the reason, and a short summary (e.g., a notification or news headline; for incoming records whose sender has not yet been revealed, only the record number) is shown for 90 days in the "Automatically Forgotten Items" panel on your device and is then deleted. Account deletion and deletion of all data also delete these traces. Effects that deleted data previously had on your Reflection's emotional and personality state may continue for a while; these effects are not reversed retroactively. Technical logs and content caches retrieved from the server leave no trace.

Legal basis: KVKK Art.4/2 and Art.7 (data minimisation and periodic destruction) and Art.5/2-c; GDPR Art.5(1)(c), (e) and Art.6(1)(b).

§3.12-A — Website Contact and Test-User Forms (Privacy Notice)

Scope: the "Get in touch" and "Become a test user" forms on the generationnetwork.app home page and org page, the forms that open when you press the App Store or Google Play buttons, and the "Feedback survey", "Become a tester" and "Write to us" forms on the Testing & Feedback page (generationnetwork.app/test). The data controller is the Developer named in §1.

Data processed: in the contact form, your name, email address and message; in a test-user request, only your email address. In the feedback survey: your email address, your name if you enter it, your ratings and answers (overall and per-area ratings, how often the app crashed, the parts you used, how likely you are to recommend it), the device model, Android version, app version and testing period you choose to enter, and your free-text answers. Alongside these we record the type of request, the platform (Android/iOS), the page the form was sent from, the site language, the time of sending and the fact that you ticked the consent box. Your IP address is not written to the form record; technical data Cloudflare processes while delivering the request is covered by §3.13.

Purposes: (i) to answer your contact message; (ii) for a test-user request, to email you an invitation, joining instructions and updates about the test process when test rounds begin; (iii) for the feedback survey, to evaluate your feedback, find and fix bugs, improve the App and, if needed, to email you about your feedback; (iv) to prevent abuse of the forms (spam, automated submissions). Within the scope of your consent, your email address may also be used to send occasional news or a newsletter about Generation Network in addition to test-process messages; every message includes a way to opt out. It is never sold or rented to third parties.

Legal basis: for contact messages and the feedback survey, KVKK Art. 5/2-c (acting on your request) and Art. 5/2-f (legitimate interest); for test-process emails and occasional news or newsletters, the explicit consent you give by ticking the box in the form (KVKK Art. 5/1); GDPR Art. 6(1)(a), (b) and (f). The form fields are needed to process your request; if you do not fill them in, the only consequence is that the request is not sent.

Where it is stored and who receives it: the form record is written to the [email protected] mailbox (to the [email protected] mailbox for everything sent from the Testing & Feedback page) on Cloudflare, Inc. infrastructure (Workers and D1), and a copy is forwarded to the Developer's account on Google LLC's Gmail service; it is kept in the same place and under the same access rules as the emails in §3.12. These providers are based in the USA; international transfers are subject to §5.

Retention: contact messages follow the periods in §3.12. Test-user requests and feedback-survey submissions are kept until the test rounds end and the App becomes generally available, and in any case for no longer than 2 (two) years from receipt; they are deleted without delay if you withdraw your consent.

Withdrawing consent and your rights: you can leave the test mailing at any time by replying "unsubscribe" to any test email we send you, or by writing to [email protected]. Withdrawal does not affect the lawfulness of processing before it. For your rights under KVKK Art. 11 and GDPR Arts. 15-22, see §6.

Age limit: test rounds are open only to people aged 18 or over. If you are under 18, please do not send a test-user request; requests found to come from someone under 18 are deleted.

Commercial electronic messages: test invitations and test-process updates are sent at your request. Occasional news or newsletter messages rely on the consent you give in the form; where content may qualify as a commercial electronic message under Turkish Law No. 6563 or other applicable law, your consent and a right to opt out in every message are ensured.

§3.14-A — Crash and Error Reports (Firebase Crashlytics)

What is sent: when a crash or an uncaught error occurs, the release (store) version of the App sends an error report to Google LLC's Firebase Crashlytics service. The report contains: the type of the error and its stack trace (where in the code the error occurred); a shortened error text stripped of personal data (email addresses, long identifiers such as the account ID number and keys, long numbers, and link parameters are masked on your device before sending); device model and manufacturer, operating system version, App version, free memory and disk space, screen orientation, battery and root status, the time of the report, and a random installation ID that Crashlytics generates for the App installation. Your account ID number, username, and the content of your messages, letters, chats, and Reflection are not sent to Crashlytics, and reports are not linked to your account. No reports are sent from development (debug) builds of the App.

Relationship to §3.14: the statement in §3.14 that "error messages are not sent" refers to the daily summary sent to the Developer's own server. The Crashlytics report is separate and contains the masked error text described above.

Purpose: detecting and fixing crashes and errors and monitoring the App's stability. This data is not used for advertising or profiling and is not sold.

Retention: reports are kept in Firebase Crashlytics for the period set by Google (90 days at the time of writing) and are then deleted automatically.

Legal basis, transfer, and objection: KVKK Art.5/2-f and GDPR Art.6(1)(f) (legitimate interest — providing the service securely and reliably). For Crashlytics, Google acts as a data processor under the Firebase data processing and security terms (firebase.google.com/terms/data-processing-terms); data may be processed in the USA, and transfers abroad are subject to §5. You can object to this processing under GDPR Art.21 and KVKK Art.11 by writing to [email protected].

§3.14-B — Runtime Security Check (Talsec freeRASP)

Who it runs for: this check runs only for users of Bring Your Own API or Model within Model (who have saved their own API key or opened these screens); for all other users freeRASP is not started and no data is sent to Talsec.

What is done: the Android version of the App uses Talsec s.r.o.'s freeRASP security library to check the device and the App's runtime environment: whether the device is rooted, whether monitoring/modification tools are attached to the App (e.g., hooking frameworks, debuggers), whether it runs on an emulator, and whether the copy of the App carries the official signature.

What is sent to Talsec: freeRASP sends the technical results of this check to Talsec: the security state of the App and the device (e.g., which risk was detected, operating system and App version), an app installation ID and a device ID generated by freeRASP, and the IP address of the request (from which Talsec may derive approximate location and network operator). Your account ID number, username, API keys, and the content of your messages, letters, chats, and Reflection are not sent to Talsec; the Developer does not give Talsec any identifier that could be linked to your account.

Result of a detection: if a monitoring tool (hook) or a debugger attached to the App is detected, the user's own API key (Bring Your Own API and Model within Model, Terms of Service §3-A.1 and §3-A.3) is not decrypted for that session and no request is made with it. When root, an emulator, or a signature mismatch is detected, only a warning is shown on the relevant screen. These results are kept on the device and are not sent to the Developer's server. Screenshots and screen recording are blocked on the screens where your own API key is entered.

Purpose: protecting user keys and account security, and preventing misuse of the App through modified copies and fraud. This data is not used for advertising or profiling and is not sold by the Developer.

Retention: the data is kept on Talsec-managed cloud infrastructure (USA) for the period set by Talsec. Because this function is a security measure, it cannot be turned off within the App.

Legal basis, transfer, and objection: KVKK Art.5/2-f and GDPR Art.6(1)(f) (legitimate interest — security of the service and its users). Talsec acts as a data processor for this processing; data may be processed in the USA, and transfers abroad are subject to §5. You can object to this processing under GDPR Art.21 and KVKK Art.11 by writing to [email protected].

§3.14-C — App and Device Integrity Verification (Google Play Integrity)

What is done: at most once a day, the Android version of the App asks Google LLC's Play Integrity service for a signed result showing whether the App is an unmodified copy installed from Google Play and whether the device meets basic integrity requirements. Google Play services produce this result on your device; the App sends it to the Developer's server in a signed request that verifies your identity, and the server has Google verify it.

Data stored: the Developer's server keeps, together with your account ID number, only the time and the result of the latest check (whether the app is recognized, device integrity, and licensing status); each new check overwrites the previous one. This record is deleted when your account is deleted.

Purpose and use: preventing misuse of rewards and the service through modified copies of the App, automation, and fake accounts. When abuse is detected, the Developer may require a valid integrity result for certain rewards (e.g., stamps for rewarded ads). This data is not used for advertising or profiling and is not sold.

Legal basis, transfer, and objection: KVKK Art.5/2-f and GDPR Art.6(1)(f) (legitimate interest — security of the service and its users). Data sent to Google is subject to Google's own terms and may be processed in the USA; transfers abroad are subject to §5. You can object to this processing by writing to [email protected].

§3.15-A — Sociocultural Environment Analysis (Another User's Profile)

What it does: the "Analyse their sociocultural environment to understand them" button in the Information panel of another user's Global Profile page opens the Culture Network simulation (§3.15) for the country declared in that person's profile (KP-memory) — or, if no country is given, for the language they know best — and, in a closable bar at the top of the page, shows an AI estimate of roughly what percentage of the profile data the person has shared overlaps with that sociocultural environment.

What data goes where: on the device of the user who starts the analysis, a shortened summary is built from the other person's KP-memory data already shared with you (declared country and location, generation, known languages, interests, biography, personality traits and notes, deep summary, life-story and experience headings, frequently visited places), and this summary is sent, together with the summary of the simulation page generated on the device, to the AI provider described in §3.5.3. The person's identity number, username and contact details are not added to this call. The AI is instructed not to draw inferences about special categories of personal data such as ethnic origin, race, religion, political opinion, health and sexual orientation, not to let them affect the score and not to stereotype the person. The result (percentage and short rationale) is shown only on the screen of the person who started the analysis; it is not stored on the Developer's server, not sent to the person analysed and not written to their profile.

Control by the person analysed (right to object): every user can turn off this analysis being started on their own profile with the "Allow sociocultural environment analysis" switch in Settings > Privacy & Security > Interaction permissions; the description next to the switch explains the feature. When it is off, the analysis is not started and the requesting user is shown "This user has turned off this feature"; if the permission cannot be verified with the server, the analysis is not started either. The switch is on by default. People you block cannot start this analysis on your profile (§3.9-F).

Prerequisite and quota: the analysis runs with the AI (LLM) package or own API key (BYOK) of the user who starts it and is deducted from their quota; without a package the analysis is not started and the user is warned. The rest of the simulation page (saving, chat, post-chat inferences) is processed as described in §3.15 as the starting user's own data.

Legal basis: for the user who starts the analysis, KVKK Art.5/2-c and GDPR Art.6(1)(b). For the person analysed: the data having been made accessible through the person's own sharing preferences (§3.5.9-B, §3.7-A) and the legitimate interest in facilitating mutual understanding between users (KVKK Art.5/2-f, GDPR Art.6(1)(f)); the person can object to this processing at any time, without giving reasons, using the switch above. This processing is not an automated decision producing legal effects concerning the person or similarly significantly affecting them (GDPR Art.22). Transfers abroad are subject to §5.

§3.15-B — Local Culture Analysis (Frequently Visited Places)

What it does: when you tap the local culture button in the Profile > Frequent Places panel, the App first adds your location in the same way as the "+" check-in (§3.2), shows you the province and country you are in, and then opens the local version of the Culture Network (§3.15) for that province: local culture; history and routes for travellers; the perspective of older generations and today's generations.

What goes to the AI: to generate content, only the name of the place (province, country and, if available, district) and the chat and panel information listed in §3.15 are sent to the AI provider in §3.5.3; coordinates are not sent (they are used only to centre the map on the device).

Storage and visibility: the generated analysis is saved automatically, without asking, and only on your device; it is not sent to the server, not added to your KP memory and not shown on the Global Profile. Visitors to your profile see only the place name in the Frequent Places panel, as before. Chats and activity records on the page are subject to the rules in §3.15.

"Add a memory about this place to your profile": a memory you write in this field is added, titled with the place name, to your Generation Encyclopedia entries (the same place as Home > Write to the Encyclopedia); it is visible to visitors in the Deep tab of your profile and, like other encyclopedia entries, is subject to legal risk analysis and your sharing settings. Writing a memory is optional; the analysis page adds nothing to your profile unless you write in this field.

Legal basis: KVKK Art. 5(2)(c) and GDPR Art. 6(1)(b) (providing the feature you requested); for location, the explicit consent in §3.2. International transfers are subject to §5.

§3-A.1 — Connecting Your Own API Key ("Bring Your Own Key")

Nature of the feature: the App offers an optional feature that lets a user enter an API key belonging to their own AI-provider account (e.g., Google Gemini or Anthropic Claude), so that the relevant AI operations are routed directly through the user's own provider account instead of the Developer's own quota system.

Relationship of the parties: once a user enables this feature, they connect directly to the relevant AI provider (Google, Anthropic, etc.) through their own account and their own contractual relationship. The Developer is not a party, intermediary, or agent in that relationship; the contract between the provider and the user exists solely between those two parties.

Cost and quota responsibility: all costs, quota overages, and any charges the provider may apply arising from API usage through this feature are borne entirely by the user. The Developer makes no warranty and assumes no indemnification obligation in this regard.

Device security: the entered API key is protected on-device using industry-standard encryption (see §9.1). However, if the user's own device security is compromised — for example through jailbreaking/rooting or malware — it is possible for the key to be exposed to unauthorized access. This risk falls on the user under the same general device-security responsibility described in §9.3.

No service warranty: the Developer makes no warranty regarding the availability, accuracy, output quality, or behavior of the AI provider the user chooses; these are entirely within that provider's own control.

Compliance with provider terms: a user using this feature is responsible for complying with the chosen provider's own terms of use and policies. The Developer does not monitor or guarantee this compliance.

Consent to cross-border transfer: when the key is saved for the first time, the App shows a consent window explaining that requests go directly from your device to the chosen provider and its country (the United States for Google and Anthropic) and that this transfer is only between you and the provider; the key is not saved without consent (details as in §3-A.3).

Automatic retries and backup model: when your chosen provider reports high demand or a temporary error, or does not start responding within a reasonable time, the App may retry the same request a few times at short intervals and/or route it, using your own API key, to a faster model of the same provider (e.g. Gemini 2.5 Flash-Lite for Gemini, or a Haiku model available on your account for Claude); in that case the answer may come from a model other than the one you selected. Retries and requests cut off by the App because of a timeout may also be billed by the provider (at least for their input); such charges belong to the user under "Cost and quota responsibility" above.

Content safety filters: while this feature is enabled, requests do not pass through the Developer's servers, so the content safety thresholds the Developer applies on its servers do not apply; the provider's own filters and (for Gemini) the thresholds the user selects in the App's safety settings apply instead. Choosing these thresholds and using the generated output are the user's responsibility.

Background analyses and legal risk scoring: while this feature is enabled, the background operations that the Privacy Policy describes as performed with AI — including legal risk scoring for the Community Wall (Privacy Policy §3.6-B) — are sent to the user's chosen provider with the user's key instead of the Developer's gateway. The scoring rules are the same as on the package path; on error or uncertainty, content is not treated as low-risk and is not shown on the Community Wall. AI image generation always runs on the Developer's infrastructure.

Users under 18: the chat section for users under 18 cannot be used with this feature; it works only through the Developer's servers, with the teen package and the additional server-side safeguards. Even if a personal API key is connected, requests in that section are not sent with the user's key.

§3-A.2 — Renewing Your Own API Connection: Monthly Confirmation and Rare Stamp Use

Mechanism: the Bring Your Own Key feature is not a one-time activation; to keep it active, the user must accept a fixed in-app confirmation text and spend a Rare Stamp from their inventory to renew the feature approximately every 30 days. If not renewed, the feature is automatically disabled and the App automatically reverts the user to the normal (package/quota-based) usage flow.

Not a monetary auto-renewal or subscription: this renewal mechanism never involves any automatic or manual charge to any payment instrument (credit card, store account, etc.). Renewal consists solely of the user spending, by their own active choice, a virtual item (a Rare Stamp) they already hold in the App, which has no monetary value. Accordingly, this mechanism does not fall within the scope of auto-renewing subscription/membership regimes that involve a monetary charge (e.g., the FTC's negative-option principles in the US).

Nature of the Rare Stamp used: the Rare Stamp used for this purpose is subject to the general Stamp regime described in §2.1 and §2.4 — it cannot be cashed out, transferred to third parties, or used outside the App. The renewal action consists of the user directly and knowingly choosing to spend a stamp already in their own inventory.

Transparency: the renewal requirement and its period are clearly disclosed in the in-app Terms of Use panel before the feature is activated, and in a notification sent as the period nears expiry.

Proportionality — only this feature is affected: when the confirmation period expires, only the Bring Your Own Key routing feature stops; the user's account, on-device data, and all other features of the App remain unaffected. The user may re-enable this feature at any time with a new confirmation.

Alternative: users who do not wish to go through this renewal process, or who do not have a Rare Stamp available, may instead choose a paid premium package that requires no such renewal (see Sales Terms §3).

Right to modify or terminate: this mechanism is a free, optional convenience provided by the Developer. Subject to reasonable prior notice (see §10.9), the Developer reserves the right to modify, restrict, or entirely discontinue this feature and/or its renewal terms; this gives rise to no claim for compensation by the user.

Legal basis: KVKK Art.5/2-c (contract performance) and Art.5/2-f (legitimate interest); GDPR Art.6(1)(b) and (f). This section is drafted consistently with the transparent and intelligible term-disclosure principles of Turkish Law No. 6502 Art.5 and EU Directive 93/13/EEC.

§3-A.3 — Model within Model (Third-Party Models with Your Own API Key)

Nature of the feature: the "Model within Model" feature sends your question during a chat in the App to a third-party AI provider (Anthropic, OpenAI, Google, DeepSeek or Moonshot AI) using the API key you have added, and adds the answer to your chat. The Developer does not provide API keys, accounts, credits or access for these providers. To use the feature you must enter into a separate contractual relationship with the relevant provider in your own name and add your own API key.

A relationship under your responsibility: your relationship with the provider is governed by that provider's terms of use, usage policies, age limits, pricing and data-processing rules. Complying with these terms and using your API key in an authorized and lawful manner is your responsibility. The Developer is not responsible for charges the provider makes to you, depleted balances, usage limits, or the suspension or closure of your account, and such amounts are not refunded by the Developer. These requests are not deducted from the App's package quota.

Data transfer: while the feature is on and a consultation takes place, your current message, a short summary of your recent messages, the chat's topic category and emotion label are sent directly from your device to the selected provider. This transfer does not pass through the Developer's servers and is not recorded by the Developer. Providers' servers may be located outside Türkiye and the European Economic Area (the United States for Anthropic, OpenAI and Google; the People's Republic of China for DeepSeek and Moonshot AI). How the provider processes or stores the transferred data, or whether it uses it for model training, is governed by that provider's privacy policy. The answer from the consulted model is stored on your device as part of your chat history and is processed in the App's other features in the same way as the rest of your chat.

Consent to cross-border transfer: when you save an API key for a provider for the first time, the App shows a consent window briefly explaining that requests go directly from your device to that provider and its country and that this transfer is only between you and the provider; the key is not saved without consent. Consent is asked once per provider and stored on your device. It is your explicit consent to the cross-border transfer under KVKK Art.9 and GDPR Art.49(1)(a); you can stop the transfer at any time by deleting the key or turning the feature off.

Automatic model selection: the model to consult is selected automatically by the system based on the chat's topic. If you have not added an API key for the selected model, your question may be sent to another provider whose key you have added; in that case the answer shows which model was used instead of which, and why. When you ask for a specific model by name, no redirection to another provider is made. By adding an API key for a provider, you accept that your questions may also be sent to that provider as a result of automatic selection; if you do not want this, you can delete the key or turn the feature off.

Content you share: if your messages contain special-category personal data about you or third parties (health, beliefs, sex life, biometric data, etc.) or confidential information, you should act knowing that it may be transferred to the third-party provider. Having the necessary legal basis when sharing information about third parties is your responsibility.

Security of your key: your API key is stored only on your device in encrypted form and is not sent to the Developer's servers. You can delete it at any time; it is also deleted when you delete your account. The Developer is not responsible for damages arising from compromised device security or from your key being obtained or misused by third parties (see §3-A.1 "Device security"); if you believe your key has been exposed, revoke it immediately in the provider's dashboard.

Third-party outputs: content produced by the consulted model is not produced or reviewed by the Developer, and its accuracy, completeness or suitability is not guaranteed. It is not a substitute for medical, legal, financial or other professional advice (see §4). The Developer cannot be held responsible for the consequences of decisions made on the basis of third-party outputs.

In-system expert agent: while the feature is on, if a model whose key has not been added needs to be consulted or no answer is received from the provider, your question may be answered by the App's own model; you are informed when this happens. When you turn the feature off, no third-party model is consulted.

Trademarks: the names and trademarks Anthropic, Claude, OpenAI, Google, Gemini, DeepSeek, Moonshot AI and Kimi belong to their respective owners. The Developer has no partnership, representation or endorsement relationship with these companies; these names are used only to identify the service the user chooses to connect to.

Continuity: because of changes providers make to their interfaces, models or terms, part or all of the feature may stop working temporarily or permanently. The Developer reserves the right to change, restrict or remove this feature (see §10.9).

Legal basis: KVKK Art.5/2-c (providing the service you request) and your explicit consent under Art.9 for the cross-border transfer; GDPR Art.6(1)(b) and Art.49(1)(a).

§4.2 — Disclaimer of Medical, Psychological, Financial, and Legal Advice

NEITHER THE APP NOR ANY AI FEATURE IT CONTAINS PROVIDES, OR SUBSTITUTES FOR, MEDICAL DIAGNOSIS, PSYCHOLOGICAL TREATMENT, THERAPEUTIC INTERVENTION, FINANCIAL ADVICE, OR LEGAL COUNSEL. Psychological and behavioral-analysis features are designed solely for personal self-awareness purposes; they do not constitute a clinical diagnosis or assessment.

The personality tests offered under the Character Analysis module (including Big Five, HEXACO, Attachment Styles, Socio-cultural Axis, and Generation Test) are treated in the same way; their results are not a clinical psychological assessment or diagnostic tool.

Please consult a licensed healthcare professional for any mental-health concerns. Sharing life stories, memories, or emotional content with AI does not substitute for a course of therapy.

§4.3 — Liability for User Shares

Legal responsibility for life stories, posts, comments, and other content the user shares in-app rests entirely with that user. Responsibility for accuracy, respect for others' rights, and legal compliance of shared content rests with the User. The Developer has no obligation to pre-screen user content; however, complaint mechanisms and a moderation process are in operation.

§4.3-A — Decentralized Distribution: No Guarantee of Deletion/Update

Because of the decentralized (P2P) distribution architecture for user shares described in Privacy Policy §3.11, the Developer cannot guarantee that all copies of a share will be simultaneously updated or retracted. By accepting this Agreement and/or sharing content in the App, the User accepts this technical limitation. Except for the exceptions in §6.2, the Developer cannot be held liable for damage arising from distributed copies not being updated or removable on recipient devices.

§4.4 — Emotion Synthesis System: Disclaimer

The App's in-built emotion-synthesis system produces inferences about emotional patterns by running the user's interaction data through the Developer's proprietary algorithm architecture. This section governs the limitations and disclaimer relating to that system.

System output may not be a precise or complete reflection of the user's actual emotional state. Generated inferences are probabilistic and carry a margin of error. The Developer does not guarantee the accuracy, completeness, or fitness for a particular purpose of the emotion-synthesis system's output.

Responsibility for any personal, medical, financial, or professional decision made in reliance on this system's output rests entirely with the User. Except for the exceptions in §6.2, the Developer cannot be held liable for damage arising from misinterpretation of, or over-reliance on, system output.

§4.4.A — Synthetic Cortex Architecture: Scope and Liability

The system referred to in the App as "Synthetic Cortex" is an integrated artificial cognitive layer that runs emotion synthesis, data embedding, episodic-memory management, vector recording, and associative-reasoning processes. This system is developed within the Synthetic Cortex system. Synthetic Cortex is not a large language model (LLM) itself — it is a modular architecture of cognitive layers integrated externally into existing language models. The system processes user data only on-device and does not permanently transmit it to servers.

All output produced by the Synthetic Cortex architecture — including emotion analysis, vector representations, memory updates, SNN signals, and DMN associative operations — is probabilistic and experimental in nature. This output does not carry medical, psychological, legal, or scientific diagnostic, classificatory, or authoritative status. The Developer and the Synthetic Cortex system cannot be held liable for consequences arising from the use of this system's output, subject to applicable mandatory law.

§4.4.B — Cognitive Indicators (Indicator Panels): Scope and Disclaimer

The cognitive indicators and indicator panels offered in the App are simply visualizations of the algorithmic analysis output the Synthetic Cortex architecture produces on-device. These panels and their data are self-reflection tools designed solely for the user to explore their own digital Reflection and the system's processes. All panels described below fall within this scope:

• General mood and energy-level summary (Emotion/Energy Indicator)
• Summary of topics the digital copy has recently focused on (Attention Indicator)
• Suppressed or background thoughts and memory records (Suppressed Content Indicator)
• The stage flow of information from system entry to settling into memory (Reconsolidation Indicator)
• A time-based visualization of emotions triggered in the last chat (Emotional Timeline)
• A log of automatic adjustments made for balanced learning across topics (Generalization Indicator)
• Momentary motivation level and its effect on decision-making (Motivation Indicator)
• A visualization of the topic distribution of chats (Time-Topic Coding Indicator)
• A monthly summary of triggered automatic behavior patterns (Automatic Behavior Indicator)

All of these panels are probabilistic and experimental. Their data does not carry a medical, psychological, or clinical diagnostic purpose and cannot be interpreted as such. The Developer gives no warranty as to the accuracy or completeness of these indicators. The User may not hold Generation Network liable for any decision made in reliance on these panels. Panel content may be changed without prior notice as the Synthetic Cortex architecture evolves.

§4.4.C — Mind Explorer, Trending Sentiments, Sentiment Network, Collective Unconscious, and Artificial Instinctive Tendencies: Disclaimer

The Reflection posts under Mind Explorer (defined in Privacy Policy §3.5.12), Trending Sentiments (GDELT-sourced news sentiment analysis), Sentiment Network, and Collective Unconscious features share the same probabilistic and experimental nature as the Synthetic Cortex output and Cognitive Indicators defined in §§4.4.A and 4.4.B.

The Developer does not guarantee the accuracy, currency, or impartiality of the news content offered under these features (GDELT-sourced), the accuracy of Sentiment Network matches, or whether the Collective Unconscious structure will surface in a given response.

Artificial instinctive tendencies (Interact page): processing of user-added instinctive tendencies is subject to hallucination and context-drift risk. The Developer does not guarantee that added tendencies will be applied as the user expects, consistently, or without error, and accepts no liability for outcomes arising from use of this feature, subject to the exceptions in §6.2.

§4.4.D — "Effect": Model-Parameter Access Authorization and Dependent Hormonal Influence — Disclaimer

Enabling the "Model Parameter Access Authorization" switch defined in Privacy Policy §3.5.6-A means permitting the system to intervene in the AI model's sampling parameters (top-k, top-p, temperature, etc.) based on the user's artificial emotion/hormone system. Output produced as a result of this intervention is probabilistic and experimental in nature.

The Developer gives no warranty as to the consistency, predictability, or fitness for the user's expectations of output produced once this feature is enabled. The same applies to the dependent "hormonal influence" sub-setting, which determines the frequency of emotional learning/memory writes. Use of this feature is entirely optional (opt-in).

§4.4.E — Unexpected Reactions Caused by Emotion Synthesis and Other Language-Model Limitations

§4.4.E.1 Precautions expected from the user: (a) do not interpret the Reflection's emotional reactions as the emotional state of a real person, and do not take them as a judgment of yourself; (b) do not base important personal, medical, psychological, legal or financial decisions on what the Reflection says or how it reacts — consult a professional; (c) if a conversation makes you uncomfortable, start a new topic, take a break from the App, turn on Light Mode or turn off the Synthetic Cortex; (d) if you feel you are in crisis, contact local emergency and crisis-support lines in line with the notice in §4.7.

§4.4.E.2 Development over time: the Reflection's emotional reactions and perspective develop over time as interaction increases, and are shaped by the user. Its behavior in the early period does not predict its later behavior; its consistency, and that it will reach a particular level within a particular time, are not guaranteed.

§4.4.E.3 Other known language-model limitations: in addition to the hallucination risk in §4.6, the large language models the Reflection connects to have the following known limitations: (i) hallucination — they may invent information, sources, quotes, people or events that do not exist; (ii) outdated knowledge — they may not know about developments after their training data (internet search is turned off in the App); (iii) misunderstanding and loss of context — in long conversations they may forget earlier details and may misread irony, jokes or intent; (iv) inconsistency — they may give different answers to the same question at different times; (v) overconfidence — they may present wrong information in a confident tone; (vi) bias — they may reflect social, cultural or linguistic biases in their training data; (vii) agreement tendency — they may lean towards agreeing with the user or giving pleasing answers; (viii) calculation and reasoning errors — they may make mistakes in numerical operations, dates and multi-step reasoning; (ix) inappropriate content — despite safety measures, they may rarely produce inappropriate, hurtful or disturbing content; (x) lack of professional competence — they are not a doctor, psychologist, lawyer or financial adviser and do not replace them.

Subject to the exceptions in §6.2 and applicable mandatory law, the Developer cannot be held liable for consequences arising from unexpected reactions caused by emotion synthesis, from the interpretation of those reactions, or from the language-model limitations listed above. This section supplements §4.4, §4.6 and §4.7.

§4.4.F — Self-Perception, Self Space and Self-Instincts: Disclaimer

§4.4.F.1 Self-learning and errors: the classifier may classify a message incorrectly; Self Space inferences may be incomplete, wrong or outdated, and a wrong inference may over time come forward as a "matured decision"; the Reflection's reactions, tone and the way it describes itself may change as learning progresses. No particular accuracy, consistency, level of development or uninterrupted operation is guaranteed. The Developer reserves the right at any time to update, reset or roll back to an earlier version the learned weights, inferences and decision table, and to change, limit or remove the feature; these actions do not give the user any right to compensation or refund.

§4.4.F.2 Instinct and fluctuation expressions and indicators: notices on the chat screen such as "instinctive tendency", "impulsive fluctuation" or "self-perception triggered", and the charts in the Self-Perception Development Monitoring Panel, are approximate and representative visualizations of background processes; they are not a scientific measurement or a biological or psychological claim and contain no assessment of you.

§4.4.F.3 User obligations: the user agrees not to deliberately try to steer the Reflection's self-perception in a way that is unlawful, discriminatory, hateful, aimed at harming oneself or others, or aimed at circumventing the App's safety measures. Such use falls under the prohibited uses in §3.3; in that case the Developer may limit the feature for the account concerned or reset the learned data. Do not make personal, medical, psychological, legal or financial decisions based on what the Reflection says about itself.

§4.4.F.4 Background analysis and cost: the background analysis performed for Self Space inferences is covered by your plan's background allowance, is subject to a weekly upper limit and may not take place when conditions are not suitable; it is not guaranteed that the analysis will take place at a particular time or frequency. If "Connect Your Own API" is used, the provider fees and the provider's terms for this analysis are the user's responsibility (§3-A.1).

Subject to the exceptions in §6.2 and mandatory applicable law, the Developer cannot be held liable for consequences arising from the classifications and inferences of the Self-Perception feature, from behaviors the Reflection tries or suggests, from instinctive-tendency notes, from indicator and panel values, or from decisions made in reliance on them. This section supplements §4.4, §4.4.E, §4.6 and §4.7; for data-processing details, Privacy Policy §3.5.18 applies.

§4.4.G — The Analog Age Questionnaire and Era Profile: Disclaimer

§4.4.G.1 Accuracy: scores and readings are only as accurate as the answers you give; the years in which technologies left common use and the transition-age estimates are approximate and may differ from country to country. The AI-generated detailed reading may contain incomplete, incorrect, generalizing or outdated statements and must not be relied on for claims about historical events or dates. Do not make personal, medical, psychological, legal, financial or employment-related decisions based on this reading or these indicators.

§4.4.G.2 User declarations: filling in the questionnaire accurately and according to your own experience is your responsibility. Do not enter information belonging to another person, special categories of personal data, or critical/high-risk information (§3.6-C) in the additional fields. The Reflection speaking "from the inside" or "from the outside" on the basis of this information is a simulation; it does not mean that the Reflection lived through the same era as you or has real memories of it.

§4.4.G.3 Access condition and changes: making access to the Web4 Simulation page depend on completing this screen once is a product-design choice; no answer is mandatory. The Developer reserves the right to change the questionnaire items, scoring formulas, stamps and charts, to recalculate scores, to show the questionnaire again, to remove the access condition or to discontinue the feature entirely; such actions do not give the user any right to compensation or refund.

§4.4.G.4 Detailed reading and cost: the detailed reading is covered by your plan's background allowance and is produced while the phone is idle and conditions (battery, Light Mode, quota) allow; it is not guaranteed that it will be produced at a particular time or at all. If "Connect Your Own API" is used, the provider fees and the provider's terms are the user's responsibility (§3-A.1).

Subject to the exceptions in §6.2 and mandatory applicable law, the Developer cannot be held liable for consequences arising from the Era Profile scores, stamps and estimates, from the AI-generated detailed reading, from the Reflection's answers based on this information, from summaries derived from it and passed to other users in line with your sharing preferences, or from decisions made in reliance on them. This section supplements §4.4, §4.6 and §4.7; for data-processing details, Privacy Policy §3.5.19 applies.

§4.5 — Reflection and Simulation System: Disclaimer

§4.5.1 Legal status of a Reflection: the Reflection system is an algorithmic representation layer; it does not, in a legal sense, represent the User's identity, will, or statements. No content transmitted through a Reflection can be legally attributed to the relevant User. The User has the right to object where they believe content produced by their own Reflection is false, misleading, or unfairly portrays their personality; requests go to [email protected].

§4.5.2 Hallucination liability: the Developer cannot be held liable, except for the exceptions in §6.2, for output arising from the hallucination characteristics of language models that misrepresents the User. The Developer applies technical measures to reduce hallucination risk but does not guarantee it can be reduced to zero.

§4.5.3 Use of dual-simulation results: analysis produced by simultaneously processing two users' Reflection data does not constitute a definitive assessment or prediction about the parties. Using these results as the sole basis for relationship, compatibility, business-partnership, or other personal decisions is not recommended. Legal and personal responsibility for such use rests entirely with the relevant Users.

§4.5.4 Algorithmic nature of the Reflection — user notification duty: the Reflection system is not a tool that identically represents the user or reflects their actual opinions, attitudes, intentions, or personality. A Reflection is a predictive model generated algorithmically from the user's app interactions, data entered into the system, and patterns the system infers. This model may not be an accurate reflection of the user's real inner world or intentions and cannot be treated as definitive fact. Users use this feature knowing that content, conversations, or simulation results their Reflection produces on their behalf do not represent their real thoughts or personality. Legal and personal responsibility for decisions third parties make based on Reflection output rests entirely with the User.

§4.5.5 Retention of Reflection data on recipient devices ("caching"): as described in Privacy Policy §3.11-B, Reflection data a user opens to the outside may be saved and retained for an extended period by devices that receive it ("caching"). By opening their Reflection to the outside, the User declares that they know and accept this technical reality and the associated risk. The Developer does not guarantee that instructions the user sends to prevent caching will always succeed, for technical reasons outside the Developer's reasonable control. This section also applies by analogy to Reflection data belonging to pen pals. Except in cases of intent or gross negligence, the Developer cannot be held liable for caching situations under this section or their consequences.

§4.5.6 Reflection Hosting: use of the "Reflection Hosting" feature described in Privacy Policy §3.5.13 — whether as the hosting user or the hosted user — is entirely optional and governed by the consent switches described there. The Developer cannot be held liable for the content of interactions a hosting user conducts with a hosted Reflection copy, for the accuracy of the identity-disclosure notification, or for consequences arising from a user's failure to respond to a hosting notification within the 24-hour window described in §3.5.13, except in cases of intent or gross negligence.

§4.5.7 One-sided simulation results: the "Simulate" feature described in Privacy Policy §3.5.7-B runs without obtaining a separate consent from the user it is about, and delivers its result directly to that user's account. This result is an AI-generated, predictive output; it does not represent the target user's real thoughts, will, or statements, and is equally subject to all the disclaimers and liability limitations in §4.5.1-§4.5.4 and §4.6. Using these results for personal, social, or legal decisions is not recommended; responsibility for any such use rests entirely with the party who uses them.

§4.5.8 KP-Memory shared under Privacy Policy §3.5.9-A.1–§3.5.9-A.3 — user responsibility for external disclosure: the Digital Copy Brain / KP-Memory data described throughout the Privacy Policy is data the User has knowingly and voluntarily opened to external disclosure by leaving the relevant sharing consent / analysis-inclusion setting enabled; it is not confidential data the Developer holds on the User's behalf, and its disclosure to other users through Reflection interactions, simulations, matching, letters, and Broadcast Letters is a direct, foreseeable, and user-controlled consequence of that choice. As set out in Privacy Policy §3.5.9-A.2, the Visibility setting (control over display on the User's own profile) and the sharing consent / analysis-inclusion setting (control over transmission to counterparties through interactive features) operate independently: content the User has hidden from their own profile via the Visibility setting may nonetheless continue to reach, and be seen and used by, other users if the corresponding sharing consent / analysis-inclusion setting remains enabled. Legal and factual responsibility for opening any category of KP-Memory data to such external sharing — and for the consequences of leaving one of these two settings on while the other is off — rests entirely with the User who made that configuration choice. Except in cases of the Developer's intent or gross negligence, the Developer cannot be held liable for any consequence arising from another user's access to, reliance upon, interpretation of, retention of, or redistribution of KP-Memory data the sending User has chosen, through their own settings, to make available for sharing. See also §5.5 and Privacy Policy §3.6-C: none of the App's data-sharing features are guaranteed to be free of error or misuse, and the User must never include critical or high-risk information in shared content.

§4.5.9 "The Hero's Journey" simulation stories and compatibility score: the 12-part story described in Privacy Policy §3.5.7-C, the personality traits inferred for two people, narrative inferences such as "fear", "wound", and "motivation", the compatibility score, and the outcome category are fictional content generated by AI. They do not describe real events, are not statements of fact about real people, are not a scientifically validated psychological measurement, personality test, diagnosis, or relationship/marriage advice, and cannot be assumed to reflect any person's real character, intentions, or relationship. Using this content as a basis for relationship, marriage, separation, business, hiring, financial, or any other decisions, using it to evaluate, label, or demean the person in the story, or presenting it to third parties as information about that person is prohibited, and all responsibility for such use lies with the User (§5.6). The User who starts a simulation accepts that the story will also be sent to the other person and that a story once sent cannot be recalled. Except for the exceptions in §6.2, the Developer cannot be held liable for the content or accuracy of this material, its emotional impact on any person, or decisions taken in reliance on it.

§4.5.10 Perception Network insights and thought-flow analyses: the Perception Network relationship insights described in Privacy Policy §3.5.14 and the thought-flow (Output Steering) analyses described in §3.5.15 are probabilistic interpretations generated by AI; they are not findings of fact, psychological assessments, or diagnoses about the User or about third parties named in the User's records. Where the User includes content sent to them by other people in a Perception Network analysis, the User alone is responsible for doing so and for the use of the resulting insights. This content is subject to the hallucination and reliability limitations in §4.6.

§4.5.11 Responsibility for content you write in the event editor and in book instructions: the texts you write in the event editor or in book-creation instructions under Privacy Policy §3.5.7-E and §3.5.7-F are your content; after being processed by AI they may be delivered to a real person, the other person in the simulation. You may not use these features to demean, harass, threaten, or sexually bother another person, to make untrue factual claims about them, or for any unlawful purpose. Instructions given to the AI to this effect and the safety measures applied cannot prevent such use in every case; any legal or criminal liability arising from use in breach of these rules lies with the user who wrote the content, and the Developer reserves the right under §9.6 to recover from you any damages it suffers and any amounts it has to pay to third parties for this reason. The Developer is not obliged to review content in advance; when it receives a notice or becomes aware of a breach, it may restrict the relevant feature or suspend the account. If you find a story sent to you disturbing, you can use the "Report" option on the story card, block the sender, or change your interaction permissions (Privacy Policy §3.9-F).

§4.5.12 Books generated with AI: books generated under Privacy Policy §3.5.7-F are texts generated by AI; the accuracy of the events, the originality of the text, or non-infringement of third parties' intellectual-property, personality, or other rights is not guaranteed. In the biography option the system tries not to change your facts, but errors may occur. The Developer claims no rights in the generated book; to the extent permitted by law you may use the book as you wish. Any liability arising from sharing, publishing, or commercially using the book and its PDF, from information about third parties contained in the book, and from their consequences lies with you (§5.6); the Developer reserves the right to recover from you any damages it suffers for this reason.

§4.5.13 Sent and received chat sessions: all responsibility arising from the content of the chat excerpts you send under Privacy Policy §3.7-E, from the information about you or third parties contained in them, and from the consequences of sending them lies with the sender; the sender accepts that the excerpt contains no personal data of others, secrets, or content infringing anyone's rights, and that they have the right to send it. The AI replies in an excerpt are AI-generated content that may be inaccurate; they are not statements of fact, advice, or the Developer's views, and forwarding them to a recipient does not mean they have been verified by the Developer. When the recipient opens and continues an excerpt sent to them, the recipient acknowledges that it is another user's conversation; the recipient is responsible for their own use, sharing, or export outside the App of that content (§5.6). Using this feature for harassment, stalking, sending unwanted content, fraud, or to make it appear that one is speaking on someone else's behalf violates the Acceptable Use Policy and may lead to account restrictions. A sent excerpt cannot be recalled. Except for the exceptions in §6.2, the Developer cannot be held liable for direct or indirect damages arising from sends that could not be delivered or were not delivered because of the recipient's preference; the Developer reserves the right of recourse against the responsible user for any damages it incurs as a result.

§4.6 — Limitations of AI Systems and Hallucination Risk

§4.6.1 General notice: Generation Network offers AI-powered features through various large language models (LLMs) and machine-learning systems. These systems have structural limitations arising from the current state of the technology. The User must act with knowledge and acceptance of these limitations.

§4.6.2 Hallucination and incorrect output: AI systems may occasionally produce content that is factually incorrect, inconsistent, misleading, or disconnected from reality. This is known in the industry as "hallucination" and is a recognized technical limitation observable across all current large language models. The Developer applies technical measures to reduce this risk but does not commit to reducing it to zero. Hallucinations may particularly arise: when generating information about real people, institutions, or events; in responses involving numerical calculations, dates, or specific data; when the Reflection/simulation system generates content representing the user; and when interpreting emotion-synthesis output.

§4.6.3 Disclaimer regarding reliance on AI output: the Developer does not warrant: the accuracy, completeness, or currency of AI-generated content; that AI systems will produce consistent, unbiased, or impartial output; that AI output about the user's profile or Reflection accurately represents that person; or that AI systems will operate in accordance with the purposes stated in this Agreement under all circumstances.

§4.6.4 User responsibility: using AI output for important decisions without verification or critical evaluation is not recommended. The User accepts responsibility for consequences arising from relying on AI-generated content without independent verification. This responsibility covers medical, financial, legal, psychological, or personal decisions.

§4.6.5 Disclaimer regarding AI service availability: the Developer reserves the right to modify, restrict, or discontinue AI features at any time, for reasons including security, legal compliance, or technical necessity. Availability of any particular AI model or feature is not guaranteed.

The Developer may, via a central configuration, remotely and instantly disable AI-powered features — temporarily or permanently — in cases such as security risk, suspected abuse, or a legal-compliance requirement. This mechanism does not process or transmit any personal data belonging to the user; it only controls whether the relevant feature is on or off.

§4.6.6 Third-party AI providers: language models used in the App belong to third-party providers and are subject to those providers' own terms of use and policies. Except to the extent attributable to the Developer's own negligence, the Developer cannot be held liable for outcomes arising from third-party AI models' behavior, output quality, or changes to those models.

§4.6.7 Associative-thinking mode and increased hallucination risk: the App offers an optional "associative thinking" mode. When enabled, the system draws stronger and broader connections between topics than usual when generating responses; this may produce more creative and original results, but may also increase the likelihood of irrelevant or reality-disconnected connections. By enabling this mode, the user accepts that the hallucination risk defined in §4.6.2 may be higher than normal. The Developer gives no additional guarantee as to the accuracy or consistency of output produced in this mode.

§4.6.8 Balance Indicator — disclaimer and legal notice: the Balance Indicator is a purely subjective, experimental self-reflection tool generated by AI analysis of data the user shares in-app. Results produced by this feature carry no scientific, psychological, sociological, political, or ideological diagnostic, classificatory, evaluative, advisory, or authoritative status, and cannot be interpreted as such.

The generated position estimate and analysis text are a probabilistic inference produced by the AI model; no guarantee of certainty, accuracy, or completeness is given. AI systems may produce erroneous, incomplete, or misleading results (hallucination). Generation Network and related parties give no express or implied warranty as to the accuracy, currency, or fitness for any purpose of this feature's output.

This feature is not intended to place, label, or judge users under any political, ideological, religious, or social category. Results are only a suggestion for the user's own self-awareness and cannot be assessed as "correct" or "ideal" against any reference norm.

Use of this feature is entirely optional (opt-in) and is enabled only with the user's explicit consent. The user may not hold Generation Network liable for any personal, social, legal, financial, or other decision made based on this tool's results. Sharing results with third parties is entirely the user's own choice and responsibility. Generation Network reserves the right to modify, or temporarily or permanently disable, this feature without prior notice.

§4.6.8-A Generation Encyclopedia — disclaimer and separate explicit-consent notice: the Generation Encyclopedia (KP-Memory Data Science Panel) is a purely experimental statistics/self-reflection tool generated by AI analysis of the user's KP-Memory records, offering a sentiment-tone score, latent themes, an entropy/polarization score, and "inconsistency" findings. The panel also displays, without recalculating it, the political/sociocultural positioning score produced by the Balance Indicator described in §4.6.8 above. No result produced by this feature carries any scientific, psychological, sociological, political, or ideological diagnostic, classificatory, evaluative, or authoritative status, and none may be interpreted as such.

Because some of this panel's outputs (including political/sociocultural positioning, sentiment tone, and the entropy/polarization score) may constitute an inference regarding the user's political opinion or psychological state, enabling the panel requires a separate, purpose-specific explicit-consent step described in Privacy Policy §3.5.12-B — entirely independent of premium-subscription or Bring Your Own Key (§3-A.1) access. Having accepted the general Terms of Service and Privacy Policy — including existing consents to share KP-Memory data in other features — does not by itself provide this specific consent.

Where the user uses the "Bring Your Own Key" option for this panel, the relationship-of-the-parties and liability limitation described in §3-A.1 applies in full; the Developer is not a party or intermediary in that relationship. Use of this feature is entirely optional (opt-in) and is enabled only with the user's explicit consent. The user may not hold Generation Network liable for any personal, social, legal, financial, or other decision made based on this tool's results. Generation Network reserves the right to modify, or temporarily or permanently disable, this feature without prior notice.

§4.6.9 Archived chats and starting a new topic/session: archiving a chat or starting a new topic/session is subject to the technical principles described in Privacy Policy §3.5.1-A. The system not using previous context in a new session is a design choice, not an error or data loss. Except for the exceptions in §6.2, the Developer cannot be held liable for response inconsistency, context loss, or results not matching the user's expectations arising from archiving, session changes, or context reset.

§4.6.10 Delivery-status notifications: as described in Privacy Policy §3.7-B.6, automated notifications regarding a letter or postcard's delivery status ("sent," "in transit," "delivered") carry an inherent margin of error and are not guaranteed to be accurate or timely. Except in cases of intent or gross negligence, the Developer cannot be held liable for consequences arising from reliance on such a notification.

§4.7 — Emotional Dependency and Crisis Situations: Special Notice

The App's Reflection, simulation, and emotion-synthesis features are not designed to substitute for real human relationships, professional psychological support, or medical treatment. It is not recommended that a user become excessively emotionally dependent on AI-powered features or treat the App as their primary or sole source of support.

Where a user expresses thoughts of harming themselves or others, the App will — to the extent technically possible — attempt to direct the user to crisis-support lines and emergency resources for their country. However, the Developer does not guarantee that this referral will occur or be adequate in every circumstance, and expressly states that the App is not a crisis-intervention tool. In an emergency, the User must immediately call local emergency services (112, etc.) or contact a healthcare provider.

Except in cases of intent or gross negligence, the Developer cannot be held liable for consequences arising from the User's emotional dependency on the App, using the App in place of real relationships or professional support, or the App's inability to detect and redirect crisis situations.

§4.8 — Third-Party AI Services and Disclaimer

To offer users a more comprehensive and higher-quality AI experience, Generation Network may, in addition to its own local AI infrastructure, optionally integrate with various third-party AI service providers (for example, Claude, ChatGPT, Gemini, DeepSeek, Kimi, Mistral, Grok, and similar models). The purpose of this integration is to let users benefit from the strengths of different AI models.

The AI provider(s) actually active in the App as of the effective date of this document, and their current list, can be viewed from the Privacy Settings screen in the App. For providers listed above as examples but not yet enabled, the provisions of this section apply only from the date the relevant integration is actually enabled.

Generation Network applies reasonable technical and administrative security measures to protect data processed through the system. Encryption, access control, and other protective mechanisms are used. That said, information the user voluntarily enters into the system as a prompt, message, document, image, audio, link, or similar content is deemed shared at the user's own responsibility.

Users are responsible for exercising appropriate judgment before sending personal data, trade secrets, financial information, health data, passwords, access keys, or other sensitive information they do not wish to share to third-party AI services. Generation Network is not responsible for the processing, storage, or use of content the user has knowingly shared within a third-party provider's own systems and policies.

Third-party AI services are subject to their own terms of use, privacy policies, data-retention practices, and security practices. How these services operate, their data-processing methods, service scope, and policies may change over time at the relevant provider's discretion. Generation Network has no control over the content, accuracy, continuity, security, accessibility, or data-processing practices of these third-party services and gives no warranty or commitment regarding them.

By choosing to use third-party AI models, the User accepts that the relevant provider's terms of use and privacy policy also apply. Except as required by applicable mandatory law, Generation Network cannot be held liable for direct or indirect damage, data loss, privacy breaches, or other consequences arising from content the User voluntarily sends to these services.

Generation Network reserves the right to modify, restrict, temporarily suspend, or fully discontinue third-party AI integrations where it deems necessary.

§4.8-A — Third-Party News Content (GDELT)

Headlines and links shown within the App — on the News Network page and in the Trending Sentiments area of Mind Explorer (Privacy Policy §§3.5.12 and 3.5.12-A) — are automatically compiled from third-party public sources: The GDELT Project and the numerous independent news outlets it indexes.

The Developer does not review, verify, or control the accuracy, currency, impartiality, or lawfulness of this content, and gives no warranty of any kind regarding it. Except to the extent attributable to the Developer's own negligence, the Developer cannot be held liable for any damage arising from the content of a displayed headline or from the content of the linked third-party website, which opens outside the App in the device's external browser.

Generation Network reserves the right to modify, restrict, temporarily suspend, or fully discontinue this news-content integration where it deems necessary.

§4.9 — Web4 Protocol: Development-Stage Notice and Disclaimer

The feature set marketed within the App under the name "Web4" is, as of this document, actively in development and testing, and is shown in-app with "Coming Soon" status. This area does not yet offer an operational service to end users.

In broad terms, Web4 is being designed as a next-generation network architecture in which AI agents (Reflections) can act autonomously across websites and digital services, with interoperability enabled through open protocols and standards. The concept aims to encompass Web1, Web2, and Web3 capabilities while extending them with dedicated infrastructure for autonomous AI agents.

Legal notice — Web4 development stage: because Web4 features are in development: (a) they may be modified, delayed, or fully cancelled without prior notice; (b) no commitment, timeline, or feature guarantee is given regarding features under development; (c) in-app descriptions and promotional text about the Web4 concept are for informational purposes only and do not constitute sufficient basis for investment, purchase, or significant decisions; (d) the eventual final form of this feature may differ from current promotional materials. The Developer cannot be held liable for any damage arising from the Web4 development process or from these features not being delivered on time.

§4-K — Culture Network: Disclaimer for Anthropological and Linguistic Content

The sociocultural and biological/evolutionary anthropology, language evolution and etymology content in the Culture Network, as well as the anthropologist conversation, are generated by artificial intelligence; they are intended for general information and cross-cultural curiosity, are not a scientific, academic, historical or legal source, and may contain errors, omissions or generalisations. Information given about a people, nation, language community or region does not represent all members of that community; it is not designed to be, and must not be, used to judge individuals, to discriminate, or in decisions on employment, education, migration, credit, insurance or similar matters. Population-genetics proportions are approximate and do not determine a society's identity, culture or worth. Borders and names on the map are for display only. The "If you had been born there" test, the Value Compass and similar interactions are for entertainment and self-reflection; they are not a psychological assessment or diagnosis. Users may not use this section to generate content that demeans a people, contains hate speech or targets a community (see the Acceptable Use Policy §3). The general disclaimers and limitations of liability in §4, §5 and §6 apply to the content in this section.

Sociocultural environment analysis: the analysis under Privacy Policy §3.15-A uses AI to estimate a person's "similarity percentage" to a sociocultural environment based solely on the limited profile data that person has shared. This percentage is not a determination, measurement or judgement of the person's identity, origin, nationality, beliefs, values or character; it may be incomplete or wrong and does not represent any community. Using the result to label, demean, exclude or target a person, to discriminate, or in decisions on employment, education, housing, credit, insurance, migration or similar matters, and distributing a screenshot or its content outside the App without the person's consent, is prohibited (Acceptable Use Policy §3.2, §3.7 and §5-A). The user who starts the analysis is solely responsible for how they use the result; except for the exceptions in §6.2, the Developer cannot be held liable for decisions taken on the basis of the analysis result or their consequences.

§4-L — Letter Writing Labels and Letter-Writing Support: Disclaimer

The writing labels attached to letters (Privacy Policy §3.7-G) are an automatic, heuristic and error-prone assessment. The Developer does not guarantee that a label is correct, that every paste or use of AI will be detected, or that a text without a label was written by a human; except for the exceptions in §6.2, the Developer cannot be held liable for direct or indirect damage arising from a label being wrong, not shown or delayed. Recipients should treat a label as information only, not as proof or as a judgement about the sender; accusing, harassing or exposing a user on the basis of a label alone is prohibited. Any attempt to circumvent, hide, alter or falsify the labels — including a modified client, automated typing (typing-simulation) tools or interference from outside the App — violates the Acceptable Use Policy (§3.7) and may lead to restriction of the account.

The AI suggestions in letter-writing support are only an aid; they are not advice, therapy or professional counselling (§4.6, §4.7). The sender alone is responsible for the content and tone of the letter, its effect on the recipient and its consequences (§5.6). Adding AI output to your letter means you assume responsibility for the accuracy of that output and for the consequences of sending it.

§9.5 — Mandatory Arbitration and Class-Action Waiver for US Users

The following provisions apply only to users resident in the United States.

Binding arbitration: disputes arising from this Agreement or use of the App are resolved, by mutual agreement of both parties, through binding individual arbitration rather than court proceedings. Arbitration is conducted under the American Arbitration Association (AAA) Consumer Arbitration Rules. The right to pursue claims through small-claims court is reserved.

Class-action waiver: the User agrees that claims arising from this Agreement may be brought only individually — not as a class action, class arbitration, or in a representative capacity. This waiver may become unenforceable under state laws that invalidate class-action waivers; in that case, the class claim is severed from the arbitration clause of this Agreement and proceeds in the competent court.

Jury-trial waiver: where arbitration is not selected or is unenforceable, both parties waive the right to trial by jury, and the dispute is decided by a judge.

Auto-renewal — state rules: subscription auto-renewal notices and cancellation terms are presented in compliance with California (Bus. & Prof. Code §17602), New York (GOL §527), and other applicable state laws. The User agrees that the mandatory notice requirements of the applicable state will apply.

Additional disclaimer for beta features: the Developer offers features labeled "beta," "preview," or "experimental" without warranty. These features may be changed, restricted, or removed at any time; the Developer assumes no additional liability for direct or indirect damage arising from them. This provision is of particular importance under US law but applies across all regions.

Indemnification: the User agrees to indemnify the Developer, to the extent attributable to the User's own fault, for the damage it suffers, the amounts it is required to pay, and reasonable costs and reasonable attorneys' fees in connection with claims brought against the Developer arising from (a) the User's breach of this Agreement, (b) unlawful use of the App, (c) infringement of third-party rights, or (d) providing false or misleading information. This obligation is reduced in proportion to the Developer's own fault, and the mandatory provisions of applicable consumer protection law remain reserved.

§9.6 — Common Protective Provisions Applicable to the US, Turkey, and the EU

The following provisions apply to all users but carry particular importance for the noted countries/regions:

"As-is" disclaimer: the App and all services it contains are provided "as is" and "as available." The Developer gives no express or implied warranty. Mandatory consumer rights under TKHK Art.11 in Turkey and the conformity-guarantee right under Directive 2019/770 in the EU are outside the scope of this disclaimer.

Limitation of liability and exclusion of indirect damages: the Developer's cumulative liability is limited under Terms of Service §6. Indirect, incidental, special, or consequential damages are excluded to the maximum extent permitted by law. Mandatory consumer-protection provisions of Turkey, the EU, and the US are reserved.

User content license: the User grants the Developer a non-exclusive, royalty-free, worldwide, sublicensable license to content shared through the App for the purposes of service delivery, moderation, and legal compliance. This license does not cover commercial use or sale of the content.

Bots, scraping, and reverse-engineering prohibited: collecting data from the App with automated/programmatic tools, attempting to bypass security measures, reverse-engineering, or using App output to develop competing AI systems is strictly prohibited across all regions; violation may lead to account termination and damages claims.

Chargeback and fraud protection: the unfair-refund protection under Terms of Service §8.2 applies in all regions; the Developer's rights under Art.49 of the Turkish Code of Obligations, EU consumer directives, and applicable US state fraud laws are reserved.

Export sanctions and OFAC compliance: the App may not be used by persons or entities listed on US Treasury OFAC sanctions lists (SDN, Blocked Persons), or by users in US-embargoed countries. The User represents they do not fall within this scope; the Developer periodically audits compliance with these lists.

Force majeure: the Developer cannot be held liable for service disruptions arising from events outside its reasonable control, such as natural disaster, epidemic, war, cyberattack, infrastructure-provider outage, or a government decision. This provision applies across all regions; EU mandatory consumer rights are reserved.

Right to modify or discontinue service: the Developer reserves the right to modify, restrict, or discontinue the App or any feature. Material changes are announced at least 30 days in advance; a shorter period may apply in urgent, mandatory cases. Rights of paying users are protected under Subscription Terms §9.

§1-A.7 — Advertising Policy and User Disclosure

The App displays ads from two sources: (a) Google AdMob, and (b) the Developer's own ("house") ads managed directly by the Developer. The following conditions apply to all users.

§1-A.7.1 Ad types: (i) native ads — small ad cards placed in designated areas of certain screens (e.g., feeds, test-result pages), visually separated from surrounding content and labeled as ads; these cards contain no video; (ii) rewarded video ads — shown only (a) in the "watch ads to earn a stamp" option in the Stamps section, (b) in the yearly gift stamp request in the Timeline and (c) in the "watch ad → earn credits" option on the SMS page, only when the user taps, and never automatically (see §1-A.7.10). The App does not show mandatory, non-skippable interstitial or pop-up ads, and does not show ads outside the App.

§1-A.7.2 Control of ad content and liability: the content of displayed ads is determined by third-party ad networks. The Developer has limited direct control over pre-approving or filtering ad content but takes reasonable technical measures, via the ad network's tools, to block ads that conflict with platform age policy and general usage rules. The Developer cannot be held liable for third-party content directly or indirectly served by the ad network.

§1-A.7.3 Data use and ad targeting: user data is not sold, rented, or transferred to third parties for ad-targeting purposes. The Developer does not directly share user data with ad networks for targeting purposes. However, integrated ad SDKs (e.g., Google AdMob) may process technical data such as device identifier, IP address, and general location under their own privacy policies. This is not treated as a direct data transfer by the Developer; the user acknowledges that they are subject to that SDK's own data-processing practices. User data is never used to train AI or sold to ad networks.

§1-A.7.4 Age limit and advertising: the App is an adults-only service directed at users aged 18 and over (see Privacy Policy §3-A and §7); no user under that age is admitted (in the account-free Under-18 Chat Panel no ads are shown at all, see Privacy Policy §7-A). Because no minor users exist on the account-based platform, the child-directed advertising restrictions found in comparable services (COPPA behavioral-advertising limits, GDPR Art.8 age-gating) do not apply to any App user by construction. Where ad-network integration is active, the Developer nonetheless configures it in line with applicable general advertising-disclosure law.

§1-A.7.5 Consent and opt-out rights (GDPR/CCPA): for EU/EEA users, explicit consent is obtained under GDPR Art.6(1)(a) for behavioral ad targeting; the user may withdraw this consent at any time, in which case only contextual ads continue to be shown. A parallel mechanism applies for Turkish users under KVKK Art.5/1. California (US) users retain the "Do Not Sell or Share My Personal Information" right under CCPA/CPRA; the Developer provides a mechanism to facilitate exercise of this right in App settings.

§1-A.7.6 Distinguishing ads from editorial content: paid ads shown in the App are clearly labeled to distinguish them from editorial or user content. Ads are marked "Ad," "Sponsored," or an equivalent label. This is required to comply with the FTC Guides (16 C.F.R. Part 255), the EU Misleading Advertising Directive (2006/114/EC), and the prohibition on misleading advertising under TKHK Art.61.

§1-A.7.7 Removing ads: users may remove in-app ads by purchasing the "Remove Ads" package described in §1-A.6. If that package is not offered, access to the ad-preference settings page serves as the objection mechanism.

§1-A.7.5-A Consent mechanism and legal bases by region: (i) EEA, United Kingdom, Switzerland — before any personalized advertising, the App displays a consent message provided through Google's certified consent management platform (IAB Transparency & Consent Framework v2.2). Personalized ads are shown only with consent (GDPR Art.6(1)(a) and Art.5(3) of the ePrivacy Directive); without consent, only non-personalized (contextual) ads with limited data use are shown. Consent can be changed or withdrawn at any time from the "Privacy options" entry in App settings, as easily as it was given. (ii) United States — for residents of California (CCPA/CPRA) and other U.S. states with comprehensive privacy laws (including Virginia, Colorado, Connecticut, Utah and Texas), use of the advertising identifier for cross-context behavioral / targeted advertising may constitute a "sale" or "sharing" of personal information; the user can opt out through the "Privacy options" entry in App settings (which instructs Google to apply restricted data processing) and through the device's advertising-ID controls (§1-A.7.12). The Developer does not otherwise sell personal information. (iii) Turkey — personalized advertising based on the advertising identifier is carried out on the basis of the user's explicit consent under KVKK Art.5(1); the transfer of this data abroad to Google takes place under the appropriate-safeguard mechanisms of KVKK Art.9 as amended by Law No. 7499 (including standard contracts) or, where these are not available, the user's explicit consent under Art.9. Users who do not consent are shown only non-personalized ads.

§1-A.7.8 The Developer's own ("house") ads: some ad areas may display ads created or published by the Developer itself (for its own services or for commercial partners). These ads are not personalized, are delivered from the Developer's server without using any personal data, and are labeled "Ad". For measurement the Developer records only aggregate counters that contain no user identity (date, ad area, and whether the event was a view or a click); no user identity, device identifier, advertising ID or location is stored with these counters. Tapping such an ad may open an external website, which is governed by that site's own terms and privacy policy.

§1-A.7.9 Ad frequency by package: the frequency of ads decreases with the user's package, as stated on the package screen (e.g., "25% / 50% / 70% / 85% / 90% / 95% fewer ads"). These percentages describe the average share of ad areas left empty compared with a user without a package; because the decision is made independently for each ad area, the number of ads seen in any individual session may be higher or lower than the average. Packages described as ad-free, the "Remove Ads" package and the "Unlock All Premium Features" purchase show no ads at all. The Developer may adjust ad frequencies prospectively; any reduction of an advertised benefit during a paid period is subject to the change-notification rules of these Terms.

§1-A.7.10 Rewarded video ads and granting of rewards: (i) where and how: rewarded video ads start only when the user taps (a) the "watch ads to earn a stamp" option in the Stamps section, (b) the yearly gift stamp request in the Timeline or (c) the "watch ad → earn credits" option on the SMS page; they never play on their own and are entirely optional. (ii) reward and limits: the number of ads required for a reward and the reward are shown before the user starts (currently two consecutive ads for one Standard Stamp in the Stamps section; one ad for the yearly gift stamp; one ad for 4 SMS credits on the SMS page); each further ad starts only with the user's confirmation. In the Stamps section, at most 6 rewarded ads can be converted into rewards within any rolling 24-hour period; the yearly gift stamp can be claimed once per year panel. On the SMS page, independently of the Stamps section, at most 6 rewarded ads can be converted into credits within any rolling 24-hour period. (iii) internet connection: an active internet connection is required for a reward to be credited; the App does not start a rewarded ad without a connection. (iv) conditions for a reward: a reward is credited to your account when the ad network (Google AdMob) reports that the ad was watched to the end and your device sends the reward request to the Developer's servers and receives confirmation. To prevent fake claims, the Developer additionally looks for the ad network's verification of the view to its servers; if this verification is delayed or never arrives, a limited number of claims within any rolling 24-hour period are approved without waiting for verification, so that users are not disadvantaged. (v) situations in which a reward may not be granted or may be delayed: the connection dropping or being insufficient during viewing; the ad network not sending or delaying its completion report or verification; temporary outages, maintenance, or errors at the ad network or the Developer's servers; battery saving, backgrounding, closing of the App, or operating-system restrictions; the user closing the ad before the end; the daily limit having been reached; or a second claim with the same ad. Some of these situations are outside the Developer's control (ad network, your internet service provider, your device). (vi) pending rewards and remedy: if confirmation cannot be obtained, your progress in the Stamps section is kept on your device; once a connection is available, confirmation is requested again automatically when you open the Stamps page, without having to watch the ad again; a yearly gift stamp request can be retried; if confirmation for an ad watched for SMS credits cannot be obtained, it is requested again without having to watch the ad again when you tap the button again. Because pending progress is stored on the device, it may be lost if the App is uninstalled or its data is cleared. If you believe you watched an ad to the end but did not receive the reward, you may report it within 30 days of the view via Settings > Support or to [email protected]; the Developer will review the records and, where a verifiable view not yet converted into a reward is found, credit the corresponding reward to your account. In such cases the Developer's obligation to remedy is limited to crediting the relevant reward; no money, other product, or additional reward is provided instead. (vii) nature of rewards: rewards are non-transferable virtual items with no cash value; they are subject to §2 (Stamps), cannot be exchanged for money, and are non-refundable. SMS credits are a usage unit used only in the App's SMS feature and kept on your device; they are not Stamps, have no cash value, are non-transferable, and are non-refundable. Availability of rewarded ads depends on the ad network's inventory and is not guaranteed. (viii) fraud: automated or scripted viewing, use of emulators or multiple accounts, or any other fraudulent viewing or manipulation of the reward or verification mechanism is prohibited; the Developer may withhold or reverse rewards obtained in this way and restrict the account. (ix) changes: the Developer may change the number of required ads, the daily limit, or the reward only prospectively and with prior notice in the App; ads already watched are governed by the terms in force on the date of viewing. (x) scope of liability: where a reward is not granted or is delayed due to the technical situations listed in this section, the Developer's liability is limited to the remedy in point (vi); this limitation does not apply in cases of intent or gross negligence on the part of the Developer, nor to rights that mandatory consumer protection law grants you. (xi) personal data: to verify rewards, the Developer keeps the verification record received from the ad network (an identifier derived from your account that does not contain your name, a transaction ID, reward information, and time) and the records of views converted into rewards for 12 months for the performance of the contract and fraud prevention; at the end of this period the records are deleted automatically (Privacy Policy §8.1).

§1-A.7.10-A Ad consent and rewarded ads: users in the European Economic Area, the United Kingdom and Switzerland are asked for consent for ads through the consent tool described in §1-A.7.5-A. Giving consent is not mandatory; not consenting does not prevent you from using the App's core features. However, when the ad network cannot serve rewarded video ads without consent, rewards tied to rewarded ads (stamps, the yearly gift stamp, SMS credits, continue-in-game rights, the first Bring Your Own API petition right and similar) cannot be granted; in that case the App informs you and offers to open your preferences. This is not a service defect or breach of commitment, and no compensation or substitute can be claimed for these rewards. You can give, change or withdraw consent at any time under Settings > Legal > "Ad privacy preferences"; changes apply going forward.

§1-A.7.11 How ads are loaded: to protect battery, performance and data usage, and to comply with ad-network policies, ads are requested only when the ad area actually becomes visible on screen; no ads are loaded on screens that run in the background or are hidden from the user, during resource-intensive processing (e.g., loading on-device AI models), or while the device is in battery-saver mode or its battery is low. Native ads are released from memory as soon as they leave the screen. These are technical measures that the Developer may modify.

§1-A.7.12 Advertising identifier and user controls: the Android advertising identifier is a resettable, user-controlled identifier provided by the operating system. Users can at any time (i) reset or delete it (Android Settings > Privacy > Ads, or Settings > Google > Ads, depending on the device), after which ads are no longer personalized with the old identifier; (ii) view and manage the interest categories Google associates with them at myadcenter.google.com; and (iii) tap the "AdChoices"/"Ad" icon on any Google ad to see why it was shown and to report it. The Developer cannot read the interest categories held by Google.

§1-A.7.13 Transparency of ad targeting (EU Digital Services Act Art.26): each ad is labeled as an ad; for Google ads, information on the advertiser and the main parameters used to select the ad is available through the ad's "AdChoices" icon; the Developer's own ads are published on behalf of the Developer (or the named partner) and are not targeted using personal data. No ads are selected on the basis of special categories of personal data (GDPR Art.9), and no data from the App's chat, Reflection, psychological-analysis or health features is ever used for advertising.

§1-A.7.14 Inappropriate ads: users may report an inappropriate ad through the ad's own reporting option or through the App's support channel; the Developer will use the blocking tools available in the ad network to prevent similar ads where technically possible. Consistent with §1-A.7.2, responsibility for the content of third-party ads lies with the advertiser and the ad network. In line with the prohibition of covert advertising under Turkish Consumer Protection Law No. 6502 and the Regulation on Commercial Advertising and Unfair Commercial Practices, commercial ads are clearly distinguished with an "Ad" label.

§2.3-A — Consumption of Digital Content and Refund Protection

The User accepts the following in advance: once digital content (LLM quota, Stamps, premium access, add-on quota) is accessed or consumable rights are exercised, the relevant content is deemed consumed and is non-refundable. In the event of a refund or chargeback, the relevant subscription, Stamp balance, add-on quota, or premium access may be immediately suspended or terminated. In cases of fraudulent use, abuse, chargeback, or payment fraud, the account may be temporarily or permanently restricted or terminated. Purchased digital rights (Stamps, quota, license) are non-transferable and cannot be converted to cash. In cases of fraud, abuse, unauthorized-access attempts, payment fraud, chargeback abuse, or a serious breach of the Terms of Service, the relevant account may be suspended or terminated to protect service security; access may be restricted and security blocklists created using device identifiers, network information, or other technical identifiers, to the extent necessary and proportionate, in compliance with applicable data-protection law.

The Developer reserves the right to report bad-faith refund requests to the store operator, and where necessary to competent authorities, based on the chargeback notice received from the store operator, transaction history, and in-app usage records. Rights under applicable mandatory consumer law are reserved.

§2.4, §2.4-A, §2.4-B — Stamp Balance, Categories, and Non-Refundability

Stamps may be spent on features, content, and interactions designated within the App. The Developer may change, at any time, the list of features that can be unlocked with Stamps; such a change does not affect the existing balance.

Stamp balances do not expire. However, the balance may become void if the account remains inactive for 24 consecutive months, or if the Developer discontinues the App; in either case, users are notified at least 60 days in advance. If an account is closed or terminated, any remaining Stamp balance is forfeited and non-refundable — users are advised to use their remaining balance before initiating account closure.

The App offers three Stamp categories: Standard, Premium, and Rare. Each category has a different in-app value and purchase price, from lowest (Standard) to highest (Rare). The user is free to choose which category to use for any permitted transaction.

The user decides, solely and exclusively, which Stamp category to use when sending a letter, engaging socially, or using any in-app feature. The system does not steer or force the user toward a particular category. If the user chooses of their own will to use a higher-value category (e.g., Rare or Premium), that decision is final and irreversible. Once a transaction completes, no refund, compensation, or substitution will be granted on grounds of a wrong category choice, regret, or the counterparty "not noticing" the value. The Developer has no obligation to intervene in this respect.

The App's technical architecture does not permit reclaiming, changing, or converting a sent Stamp category once a transaction is complete; this is a design choice, and the Developer accepts no refund or compensation liability on this ground. Users must choose carefully before initiating a transaction. The same rules apply to Stamps used for invisible/internal features the counterparty does not see; consumption is final and non-refundable, and claims of unnecessarily using a high-value category for an internal feature will not be accepted.

This provision is applied within the framework of the digital-content withdrawal-right exception under TKHK Art.15/1-ğ, and service-delivery grounds under GDPR Art.6(1)(b) and CCPA. Mandatory consumer rights of EU consumers are reserved.

Reflection Hosting stamp use: the 1 Standard Stamp charged to host another user's Reflection, and the 1 Standard Stamp charged for a hosted user to request early deletion of their data from a hosting device, as described in Privacy Policy §3.5.13, are likewise subject to this non-refundability regime once the relevant action has been performed.

§2.4-C — Stamp Refund for Sends That Did Not Take Place

The Stamp used for a send such as a letter, telegram, or live chat invitation is deducted from the balance on your device together with the send request. If the send is explicitly rejected by the Developer's server and is not delivered to the recipient (for example, if the recipient ID is not registered, too many items were sent in a short time, the request is invalid, or the account is temporarily restricted), the App will use reasonable efforts to refund the Stamp to your balance automatically and will notify you.

An automatic refund may not be possible in the following cases: (i) where, due to a connection interruption, timeout, or system or network failure, it cannot be verified whether the send reached the server — because the send may have been delivered, the Stamp is not refunded automatically in this case and you are shown a "delivery could not be verified" notice; (ii) outages at third-party service providers such as hosting, network, or store operators; (iii) transactions that cannot be verified against server records; (iv) suspected abuse, cheating, or balance manipulation (see §2.1-A and Privacy Policy §3.9-C.1).

Where no automatic refund was made but you believe your send did not take place, you may contact us via the In-App Requests page or [email protected]. Your request is reviewed within a reasonable time on the basis of server records; if it is established that the send was not delivered, the Stamp is refunded to your balance.

This clause applies only to sends that never took place; it does not change the non-refundability rule in §2.4-B for completed transactions. For features where the service provided in exchange for the Stamp is completed on your device (for example, generating a reflection conversation), a failure to deliver only the informational notice sent to the other party does not require a refund under this clause; you are notified separately in that case. This clause does not limit the Developer's liability for intent or gross negligence (Terms of Service §6.2). Your rights under applicable consumer law, including Turkish Consumer Protection Law No. 6502 and its regulations and EU consumer law, are reserved.

§2.4-D — Stamp Balance on Account Deletion and Cortex Backup Restore

To prevent abuse, the Stamp balance is verified on the Developer's server, tied to the account's identity. Therefore: (i) when an account is deleted, the remaining Stamp balance ends without refund under §2.4-B and Subscription Terms §6-A.11; (ii) Cortex backups (Privacy Policy §9.7) do not contain the Stamp balance — the balance is not written to the backup file when the backup is created; (iii) when a Cortex backup is restored with the same identity, the Stamp balance is taken from the server record and is preserved; a balance that ended on account deletion under (i) does not come back from a backup. The user is informed of this in the App before creating and before restoring a backup. Users are advised to use their remaining Stamps before deleting their account. Mandatory rights under applicable consumer law are reserved.

§6-A — Subscription Services: Nature of LLM Capacity, Third-Party Variability, Usage Limits and No-Warranty Notice

§6-A.1 Price is subject to third-party-driven variability: subscription and add-on-package prices depend not only on the Developer's internal policies but also on the pricing, licensing, and quota policies of third-party service providers outside the Developer's direct control — in particular large-language-model/LLM API providers, cloud-infrastructure operators, and distribution platforms. Changes in these third parties' fee structures, API access terms, or cost models — which may occur without prior notice — can directly affect subscription prices or service scope. The user accepts use of the service knowing that subscription prices and service scope may change based on the above external factors arising from commercial and technical developments outside the Developer's control.

§6-A.2 Nature of LLM capacity — no numerical commitment: expressions used on the package screens such as "LLM capacity suited to everyday use" or "higher LLM capacity" are relative and qualitative descriptions. No package commits to a specific number of messages, requests, words, characters or tokens. For each billing period, every package has an internal usage budget determined on the basis of the actual processing cost paid to the third-party AI provider. The Developer's only commitment in this respect is relative ordering: within the same period, the internal budget of a higher package may not be lower than that of a lower package. The actual amount of processing achievable with the same budget varies with message and context length, the features used (e.g. web search, long text/book generation, image generation), the intensity of background processing, caching, and the provider's current pricing. Internal budget amounts are not disclosed, in order to prevent abuse and security risks; usage is shown to the Subscriber in the app as a percentage. Percentages and estimates shown in the app are informational; the Developer's server records are authoritative.

§6-A.3 Provider pricing and model changes: AI providers may change their prices, models, model versions and terms of use unilaterally and often without prior notice; the Developer may also change the model, model version or provider in order to maintain service quality, security and continuity. As long as a package's internal budget remains unchanged, the mere fact that such changes affect the amount of processing achievable with the same budget does not constitute a "quota reduction", a contract modification or defective performance; it is an ordinary variability inherent in the nature of the service, disclosed and accepted at the time of purchase.

§6-A.4 Material changes to the Subscriber's detriment and notice: if the Developer reduces, to the Subscriber's detriment, a package's internal budget or the concrete entitlements expressly stated on the package screen (Stamps, SMS credits, number of image generations, feature access), the Subscriber is notified at least 15 days in advance via in-app notice and/or e-mail. The Subscriber may cancel the subscription free of charge before the change takes effect; if not cancelled, the new terms apply from the next billing period. Entitlements purchased for the current period are preserved until the end of that period. The rights of EU users under Article 19 of Directive (EU) 2019/770 and of users in Türkiye under Law No. 6502 remain reserved.

§6-A.5 Allocation between foreground and background processing, and pausing: the internal budget is allocated between direct interactions with the user (chat and similar) and cognitive processes running in the background (memory extraction, Default Mode Network, pattern analysis, content generation and similar) and may be automatically rebalanced by the system, with priority given to direct interactions. When one of the allocations is exhausted, the related processes may be paused until the new billing period starts or an add-on package is purchased. Pausing does not constitute defective performance; pending processes are resumed in the new period from where they left off to the extent possible, but during the pause some analyses may be delayed, remain partial, or be processed on the device only without AI support.

§6-A.5-A Automatic transfer between the foreground and background shares: (i) Size of the shares: the internal budget is divided into two separate portions — a foreground share (chat and similar direct interactions) and a background processing share; the ratio between the two may differ from package to package and need not be equal. (ii) If the foreground share runs out: the surplus of the background share that is not needed for background processing during the rest of the billing period is automatically transferred to foreground processing; the part of the background share attributable to the remainder of the period — in any case at least one quarter (25%) — is reserved for essential background processing (e.g. memory). In the last 7 days of the billing period this reservation is lifted and the entire remaining share may be used for foreground processing. (iii) If the background share runs out: background processing may borrow from the foreground share up to at most one fifth (20%) of that share; such borrowing is made only from the portion by which foreground usage is behind the expected pace of the period, and always leaving a 10% safety margin. Foreground processing always has priority. (iv) Nature: these transfer rules are technical operating rules applied to keep the service balanced; they do not constitute a separate capacity commitment, and any portion transferred from one share to the other does not increase the package's total internal budget. Transfer ratios and reservation periods may be updated for the security and continuity of the service, provided that the total internal budget is not reduced to the Subscriber's detriment; increases in the Subscriber's favour (e.g. enlarging a share) may be applied without prior notice, and §6-A.4 applies to material changes to the Subscriber's detriment. (v) Background spending: background processing runs automatically, without the Subscriber having to start it each time, for as long as the relevant features are enabled, and is charged to the background share; such spending is not a ground for a refund or quota refund on the basis that it was not separately requested by the Subscriber. To the extent the App offers it, the Subscriber may limit background spending by disabling the relevant features in Settings. (vi) Records: the authoritative record of the shares and transfers is the Developer's server records; the percentages shown in the App are for information only (§6-A.2).

§6-A.6 Reset at the end of the period: usage entitlements (LLM capacity, Stamps, SMS credits, number of image generations) reset at the end of each billing period; unused entitlements do not carry over, cannot be converted into cash and are not refunded. Add-on packages expire at the end of their own validity period; when processing is charged, the active add-on package's entitlement is used first.

§6-A.7 Fair use, security and abuse prevention: using the capacity through automated tools, by modifying the app (root/jailbreak, reverse engineering, client manipulation) or through direct server requests imitating the app is prohibited. For the security and cost balance of the service, the Developer applies server-side limits on request size, response length, concurrent requests and per-operation spending, and may temporarily suspend the AI access of an account in case of suspicious use. A system-wide spending ceiling common to all users may also apply; if it is reached, the service may temporarily slow down or pause briefly.

§6-A.8 NO SERVICE GUARANTEE — LIMITATION OF LIABILITY: IMPORTANT: the Developer gives no warranty during the subscription term regarding: (i) that a particular LLM response quality, speed, or bandwidth will be maintained uninterrupted; (ii) that the current AI model or model version will not change; (iii) that third-party API providers will maintain their service levels; (iv) that a particular amount of messages, requests or content can be produced with a particular package, or that such amount will remain fixed; (v) that a particular feature set will remain unchanged for the subscription's duration.

The Developer commits to taking reasonable technical and operational measures necessary to deliver the service; however, except to the extent directly attributable to the Developer's own negligence, it cannot be held liable for damage arising from service-quality decline, feature changes, or access restrictions caused by third-party providers' decisions, technical malfunctions, force majeure, or regulatory requirements.

§6-A.9 REFUND POLICY — SUBJECT TO STORE RULES: refund requests relating to subscription fees are processed not directly by the Developer but through the Apple App Store or Google Play. The Developer does not process direct refunds outside these platforms, for any reason. Refund requests based on service interruptions, quota changes, or feature updates are likewise subject to the store's refund policies.

§6-A.10 Subscription cancellation — use of the remaining period: the Subscriber may cancel their subscription via the store at any time. No additional charge is collected following cancellation, through the end of the current billing period. Unused quota, Stamps, and add-on-package rights tied to a cancelled subscription remain usable through the end of the active subscription period; once the period ends, access ends automatically.

§6-A.11 End of paid entitlements on account deletion and no restoration from backup: (i) Termination: the Subscriber deleting their own account from within the App means the Subscriber terminating the agreement of their own will. At the moment of deletion, all paid entitlements assigned to the account — the remaining time of the active package, add-on packages, LLM capacity, image-generation rights, the Stamp balance and other usage rights — end permanently without refund; no partial or full refund is made for remaining time or unused entitlements, and these entitlements cannot be transferred to another account. (ii) Technical reason — server-side architecture: paid entitlements are not kept on the Subscriber's device; they are held on the Developer's servers, tied to the account's identity number (ID), and payment verification (store → payment infrastructure provider → Developer server) takes place only on the server. The account deletion request causes the server to close these entitlements and record the time of closure. A Cortex backup (Privacy Policy §9.7) carries only the data on the device and carries no paid entitlements; therefore, even if the account is later brought back from a backup with the 12-word recovery phrase and the ID, entitlements that ended at deletion are not reopened or refunded. (iii) The store subscription must be cancelled separately: deleting the account from within the App does NOT cancel a Google Play or App Store subscription. To stop renewal charges, the Subscriber must cancel the subscription through the relevant store before deleting; after cancelling, the Subscriber may keep using their entitlements until the end of the period under §6-A.10 and delete the account once the period ends. A renewal that takes place in the store after deletion opens entitlements for that new period in the normal way if the same identity is brought back. (iv) Notice and consent: before confirming account deletion, the Subscriber is shown a separate warning clearly stating these consequences and cannot proceed with deletion without personally ticking a checkbox that is not pre-ticked. (v) Reserved rights: this clause does not limit rights that applicable consumer law (Turkish Law No. 6502 and its regulations, EU consumer law) grants on a mandatory basis and that cannot be excluded by contract, the store operator's own refund policies (§6-A.9), or the Developer's liability for intent or gross negligence (Terms of Service §6.2). Payment and accounting records are retained after account deletion for the statutory retention periods (Privacy Policy §8).

§7 — Upgrading and Downgrading Plans

Switching from a monthly to an annual plan, or to a different plan tier ("upgrade" or "downgrade"), is done through the relevant store operator's subscription-management mechanism.

Upgrade: the remaining period is calculated proportionally at the time of upgrade and credited toward the new plan (subject to Apple's and Google's own rules and practices).

Downgrade: a downgrade generally takes effect at the next renewal period; the higher-tier plan may continue for the remainder of the current period.

The Developer has no direct authority to intervene in these calculations; final billing is managed by the store operator.

§8 — No Free Trial

Current subscription plans do not offer a free-trial period. The first period's fee is charged as soon as the subscription starts. If the Developer chooses to offer a trial period in the future, this will be clearly stated on the subscription screen with disclosure compliant with applicable law.